liangzid/robench2024b_all_setq-bioSCP-s · Datasets at Hugging Face

shuffled_text stringlengths 267 3.59k	A stringclasses 6 values	B stringclasses 6 values	C stringclasses 6 values	D stringclasses 6 values	label stringclasses 4 values
A: The PCAM dataset was downloaded from the original website (https://github.com/basveeling/pcam). All images have a size of 96 x 96 pixels, in three colorsB: The training set has 262,144 images (80 % of the total), the validation set has 32,768 images (10 %) and the test set also has 32,768 images (10 %). All datasets have a 50/50 balance between positive (tumor present) and negative (tumor absent) samplesC: The patches of 96 x 96 pixels images were automatically extracted from the CAMELYON dataset [13]. For each image, a positive label indicates that the 32 x 32 pixel center of the image contains at least one pixel annotated as tumor tissue (Figure 1).	CAB	ABC	ACB	CAB	Selection 2
A: By definition, an evolutionary stable condition is surrounded by unfit traits, at least within an α𝛼\alphaitalic_α-radiusB: This form of a fitness landscape is referred to as a fitness valley and has been studied in a special case in [8]C: Based on this, we introduce a measure for the stability of a coexistence equilibrium, connected to the width of the surrounding fitness valley.	BAC	ABC	CBA	BAC	Selection 2
A: We first compute the virus-free equilibrium and endemic equilibrium after establishing the basic reproductive numberB: Subsequently, we extend our analysis to the stochastic counterpart, determining the basic reproductive number in the stochastic framework. This stochastic reproductive number becomes instrumental in discerning the dynamics of coronavirus extinction and persistence, contributing valuable information for disease control strategies.C: In this section, our objective is to derive sufficient conditions that elucidate the extinction and persistence of the disease, providing novel insights into the control of COVID-19	ABC	BAC	BCA	ABC	Selection 3
A: In Wang et al. (2017), persistent homology was shown to outperform topographic power maps. In (Yoo et al., 2017), center persistency was shown to outperform the network-based statistic and element-wise multiple corrections. In Chung et al. (2023b), persistent homology based clustering is shown to outperform k𝑘kitalic_k-means clustering and hierarchical clustering.B: In Lee et al. (2011, 2012), persistent homology was shown to outperform eight existing graph theory features, such as clustering coefficient, small-worldness, and modularity. Kuang et al. (2019) showed persistent homology-based measures can provide more significant group difference and better classification performance compared to standard graph-based measures that characterize small-world organization and modular structureC: In Chung et al. (2017b, 2019a), persistent homology was shown to outperform various matrix norm-based network distances. In Wang et al. (2018), persistent homology was shown to outperform the power spectral density and local variance methods	CBA	ACB	CAB	BAC	Selection 3
A: In this note, we propose a multiscale model for glioma development and spread which interconnects the dynamics of glioma cells, vasculature, and vascular endothelial growth factors. Glioma cells mainly spread according to the anisotropy of brain tissue or moving along the blood vessels therein [17, 38]. In order to provide nutrient and oxygen supply to its growing cell population, the tumor produces and releases in the extracellular environment growth factors, such as VEGFs, that attract endothelial cellsB: The resulting angiogenesis boosts the growth of the neoplasm, leading to an abundantly vascularized and progressed tumor mass. Standard treatment usually comprises chemo- and radiotherapy, in a concurrent or an adjuvant manner. Thereby, both tumor and normal tissue are affected, although to a different degreeC: A necrotic region is formed within the tumor mass, due to impaired tumor-associated vascularization. We also consider here the effect of an anti-angiogenic drug (e.g., bevacizumab [28]), aimed at reducing the affinity of VEGF receptors on ECs toward their ligand released in the extracellular space by the tumor cells. This therapeutic approach affects the EC tactic motility by reducing their bias towards (increasing gradients of) VEGFs and impairing their proliferation, which, in turn, leads to degradation of glioma through diminished nutrient supply. Starting from the subcellular level of interaction between cells, VEGFs, and tissue, we set up the corresponding kinetic transport equations (KTEs) for glioma cells and endothelial cells, and perform a (non-rigorous) parabolic limit to deduce the macroscopic system of reaction-advection-diffusion PDEs for the involved cell populations: glioma and endothelial cells. These equations are then coupled to the evolution of VEGF concentration, healthy tissue, and necrotic matter, which are considered directly on the macroscopic scale.	ABC	BCA	ACB	ACB	Selection 1
A: For comparison, two other types of brains are also included. Purple points represent lissencephalic brains, with hemisphere sizes measured as depicted in Fig. 3B: Red points denote quasi-gyrencephalic brains, measured as shown in Fig. 4 and Fig. 5. Data are collected from academic publications and www.brainmuseum.org. For species with multiple images, each image is measured individually. When possible, brain sizes are measured along the lateral axis, encompassing both hemispheres. C: Plot of gyral size versus brain size for gyrencephalic brains. Blue and green points indicate gyral sizes measured using different methods, as illustrated in Fig. 1 and Fig. 2, respectively	ABC	CAB	ACB	BCA	Selection 4
A: Often, novel drugs may not demonstrate efficacy in clinical trials despite thorough preclinical safety testingB: Detecting synergistic effects among approved drugs holds clinical significance, as drug repurposing can streamline the expensive and lengthy process of developing new drugsC: In the future, integrating in silico drug synergy discovery with ex vivo testing of selected anti-cancer drug combinations in patient-derived 3⁢D3D\mathrm{3D}3 roman_D organoid models could significantly enhance the clinical translatability of this approach.	CBA	ACB	ACB	BAC	Selection 4
A: 2F), there is an optimal number of chromosomes for evolving novel traits (FigB: 2B).C: We found that the skewness of the distribution of genotypes coded in each chromosome is essential to determining the probability of evolving new traits. Because skewness peaked with a finite number of chromosomes (Fig	BCA	ABC	CBA	CAB	Selection 1
A: The gradient component sum to zero along any cyclesB: The curl components are zero for edges that are not a 2-simplex boundary and the entries sum to zero around each node. The harmonic component sums to zero around each node, and it also sums to zero along each 2-simplex. We tested the topological equivalence of female brain networks and male brain networks using the Wasserstein distance (5).C: We further determined if we can detect topological differences in the decomposed components (Figure 2)	BCA	CAB	ABC	ACB	Selection 1
A: preprint arXiv:2210.05819).B: Others are modifications, such as models with varying population sizes [Mohle2002, KajKrone2003, Freund2020] or diploid reproduction [MohleSagitov2003, BirknerLiuSturm2018], using heuristics close to those of [MohleSagitov2001]. Finally some works use different techniques such as duality, e.gC: in the case of populations suffering recurrent bottlenecks [CasanovaMiroJegousse2020] or overlapping generations (Casanova et al	CAB	ACB	CBA	CBA	Selection 1
A: Noise introduces significant interference into the model, and we aim to ascertain whether attention consistency still holds under noisy conditions. By employing these strategies, we aim to ensure a comprehensive evaluation of our model across different learning scenarios. B: To comprehensively validate the effectiveness of attention-consistency learning, we introduced a noisy model, FedAvg-N, as another baselineC: Our experiments encompassed three distinct settings: (1) local learning: Each center independently trained its data; (2) global learning: aggregating datasets from all centers, along with training a multicenter centralized model; (3) federated learning: Training on data from all centers without sharing datasets, focusing solely on exchanging model parameters	BAC	CBA	ACB	CAB	Selection 2
A: In the experiments, we will demonstrate the importance of taking the alignment of data into consideration by comparing our method to these baselines.B: Solving DSBs is a subject of significant interest in recent years and has flourished in a number of different algorithms (De Bortoli et al., 2021; Chen et al., 2022a; Vargas et al., 2021; Bunne et al., 2023; Liu et al., 2022a)C: However, all these previous approaches focus on unaligned data, and therefore the methodologies all rely on IPF and are hence drastically different from ours	BCA	ACB	BCA	CAB	Selection 4
A: Here our aim is to investigate in the context of a toy model the consequences of density dependence that only affects development directly (fertility is affected indirectly, since it depends on the developmental stage of the individual)B: We do so for a one-dimensional i-state (i.e., the variable capturing the relevant differences among individuals ‘lives’ on the real line), so for an i-state space that comes equipped with an order relationC: In fact we shall assume that the presence of ‘larger’ individuals has a negative impact on the growth rate of ‘smaller’ individuals (as a motivating example one might think of trees and shading, with the i-state interpreted as ‘height’; but please note that we ignore spatial structure and that, consequently, the model is but a caricature). For the incorporation of space into physiologically structured population models see [27]. For an alternative approach to hierarchically structured models see [25].	CBA	ACB	BAC	ABC	Selection 4
A: Specifically, the procedure on the twin lattice was more than five times faster, requiring between 16%percent1616\%16 % and 20%percent2020\%20 % of the time required by the procedure on the model inclusion lattice. Furthermore, the proper implementation of the coherence principle allowed us to fit a much smaller number of models, ranging between 37%percent3737\%37 % and 54%percent5454\%54 % of the models fitted under the naive implementation of principle of coherenceB: The point of main interest is the comparison in terms of efficiency, that we quantify with respect to the average execution time and the average number of fitted models. These can be found in the last two columns of Table 2 and, furthermore, the growth rates of these measurements are displayed in Figures 6 and 7. We can see that the procedure on the twin lattice is considerably more efficientC: It is also interesting to notice that the latter proportions appear to decrease as p𝑝pitalic_p increases. Table 2 also gives the average values over the 20 samples of the positive predicted value, the true-positive rate and the true-negative rate, both for the edges and for the colour classes of the selected graphs. These have satisfying values and there are not relevant differences between the two procedures, thereby showing that the increase in efficiency is not achieved at the cost of a lower level of performance of the selected model.	CBA	BAC	CBA	BCA	Selection 2
A: However, the current study is limited to single institutional data. In the future, we would like to explore the applicability of the method for multi-institutional data using domain adaptation techniques. B: This paper proposed a tensor learning-based pipeline for PAWP classificationC: We demonstrated that: 1111) tensor-based features have a diagnostic value for PAWP, 2222) the integration of CM features improved the performance of unimodal and bi-modal methods, 3333) the pipeline can be used to screen a large population, as shown using decision curve analysis	BAC	CBA	CBA	CAB	Selection 4
A: Cat refers to concatenation, KP refers to Kronecker product. All omics and multimodal baselines were trained with the Reactome and Hallmark pathway sets. B: Best performance in bold, second best underlinedC: Table 1: Results of SurvPath and baselines in predicting disease-specific patient survival measured with c-Index (at 20×\times×)	ACB	CBA	BCA	ABC	Selection 2
A: Most notable is the effect of such systems to smooth and provide an additional degree of control over external noise, consequentially increasing resilience and robustness. The inclusion of feedback and feedforward loops can enhance this effect, providing systems with additional degrees of control and contributing to so-called perfect adaptation [46, 45]B: Such results potentially explain the complexity in the replication network structure seen in some viruses; for example, in the human adenovirus [37]. Indirectly, these loops (and by extension, more complicated network structures) provide systems with the ability to tune the first passage time distribution, potentially yielding an optimal lysis time [47, 13]. While we restrict our analysis to a single non-local feedback or feedforward, future work may study more general optimal network structures informed by more specific biological problems.C: Such loops provide a potential explanation for out-of-order progression in some systems, for example, whereby viral replication does not occur as a unidirectional process [2]. Our work demonstrates that feedback loops could yield a fitness advantage through more favourable statistical properties of viral load compared with perfect progression through the replication cycle	CAB	CBA	ACB	BCA	Selection 3
A: Elman networks, more commonly known as vanilla recurrent neural networks (RNN), attempt to introduce the concept of a time-dependent dynamic memory [16]. The idea is to make predictions about inputs based on contextual informationB: Context-based predictions can be done for four input-output schemes: one-to-one, one-to-many, many-to-one, and many-to-many. One-to-one models are a variation of a classic neural network, one-to-many models are best for image caption generation, many-to-one models are best for sentiment analysis, and many-to-many models are best for translation or video frame captioningC: Fig. 1 is an example of the basic structure of a vanilla RNN.	CBA	BCA	ABC	BAC	Selection 3
A: 1). The three regimes correspond to the network topology updates being much faster (annealed regime), comparable (intermediate regime) or much slower (quenched regime) than the spread of the disease. B: The analysis of epidemics on time-variant networks is often under the assumption of timescale separation: either the network changes significantly faster than the spread of the disease or the epidemic evolves significantly faster than the topology updates of the networkC: Timescale separation of the network and the epidemic results in three regimes (Fig	CAB	BCA	BCA	ACB	Selection 1
A: We explore features of covers arising from networks, and characterise many of the familiar classes in terms of properties of their associated covers. It is to be hoped that encoding network properties in the properties of sets of sets will enable some new directions to be pursued in studying phylogenetic networks. B: They have been shown to correspond to a set of covers of finite sets that satisfy a property called “expanding”C: The class of labellable networks contains many commonly studied classes	ABC	ABC	CBA	CAB	Selection 3
A: Suppose we are given a sample from the selected locus at the present time t=0𝑡0t=0italic_t = 0, and that we know the allelic types of the sample but we do not know how the sample was produced. What is the genealogy of the sample? This question was answered by Barton et alB: For samples from a population at stationarity as in Section 1, Barton et al. (2004) proved that this could be done rigorously starting with a Moran model with finite N𝑁Nitalic_N then passing to the diffusion limit. Barton and Etheridge (2004) explored some properties of gene genealogies under this model, and Etheridge et al. (2006) used the same idea to describe genetic ancestries following a selective sweep. C: (2004), who modeled the ancestral process using the structured coalescent with allelic types as subpopulations. The structured coalescent can be a model of subdivision with migration between local populations (Takahata, 1988; Notohara, 1990; Herbots, 1997) or a model of selection with mutation between allelic types (Kaplan et al., 1988; Darden et al., 1989)	CBA	ACB	BCA	ABC	Selection 2
A: MIXALIME also provides machinery to test for the differential allele-specificity between two sample groups (i.eB: control and test)C: We employ Wald or likelihood-ratio test (LRT) to see if there is a difference in parameters estimates between two groups:	BCA	BAC	ABC	BAC	Selection 3
A: fully susceptible (σ⁢(a)=1𝜎𝑎1\sigma(a)=1italic_σ ( italic_a ) = 1)B: If TR=∞subscript𝑇𝑅T_{R}=\inftyitalic_T start_POSTSUBSCRIPT italic_R end_POSTSUBSCRIPT = ∞ a.s., individuals are permanently immune following an infectionC: Depending on the distribution of TRsubscript𝑇𝑅T_{R}italic_T start_POSTSUBSCRIPT italic_R end_POSTSUBSCRIPT, we can recover either the SIR, SIS or SIRS model	CBA	BCA	ACB	BCA	Selection 3
A: In particular, the analytical species abundance distribution derived from the DMFT follows the Gamma distribution, a widely utilized probability distribution in macroecology [32, 1]B: In other words, due to the non-linear nature of the corresponding Fokker-Planck equation and known pathologies in the GLV model (also observed in the quenched case), the dynamics may not converge to the stationary solution, leading to divergent trajectories.C: Again, similar truncated fat-tailed distribution has been recently shown in the chaotic phase [36] and in the strongly interacting limit [5, 41] of the QGLV with immigration. We have successfully obtained the phase diagram for the case of annealed white noise (AWN), and numerical simulations for the case J⁢(x)=x𝐽𝑥𝑥J(x)=xitalic_J ( italic_x ) = italic_x have revealed the potential for unbounded growth when the initial conditions possess large values, despite the existence of an analytically stationary solution	BAC	BAC	ACB	CAB	Selection 3
A: Transcription—the synthesis of RNA—is typically modelled as a multistep process in which the gene switches between multiple states before it eventually produces an RNA moleculeB: Depending on the level of details, transitions between gene states may reflect individual biochemical events, such as binding of transcription factors and RNA polymerase at the promoter, or more phenomenologically, a combination of these events that results in the gene being either active or inactiveC: Once the RNA is produced, it goes through a series of steps until it is eventually degraded. In the Markovian setting, these steps can be described by the following reaction scheme	CBA	ABC	BAC	CAB	Selection 2
A: Ho et alB: (2018) calculate transition probabilities for the birth/birth-death process—a restricted bivariate case where the death rate of one type vanishes, but rates may be otherwise nonlinear. Xu et alC: (2015) use branching process approximations of birth-death processes and generating-function machinery (Wilf, 2005) for moment estimation.	CAB	BAC	ABC	BCA	Selection 3
A: Where the similarity represents the highest alignment score between the amino acid sequences of the test and train sets using BLAST identityB: Furthermore, in Figure 2 we report the performance of all Prot2text models with respect to different similarity thresholdsC: We observe that for test proteins with low similarity scores with the train set (between 20% and 30%) and for proteins with no counterpart in the train set, the Prot2TextMEDIUM is the dominant one while for higher similarity scores Prot2TextLARGE performs better.	BCA	BAC	BCA	CBA	Selection 2
A: strength [cfB: Fig. 3(b) and discussion after Eq. (31)]C: Note that the saddle point 𝑸∗⁢(𝑫=𝑫¯)superscript𝑸𝑫¯𝑫\boldsymbol{Q}^{*}\left(\boldsymbol{D}=\overline{\boldsymbol{D}}\right)bold_italic_Q start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT ( bold_italic_D = over¯ start_ARG bold_italic_D end_ARG )	BCA	CBA	ABC	CBA	Selection 3
A: As shown in Figure 7, changes in the shape of the reconstructed volumes indicate a large flexible motion in the continuous conformational space. The corresponding predicted volumes and the distribution of predicted latent z𝑧zitalic_z along the first two principal component axes are similar to that from the literature [30].B: For experimental datasets, the distribution of noise is often more complex than simulated datasetsC: Based on our simplified noise estimation method (Equation 9), the SNR of spliceosome datasets is around −11.711.7-11.7- 11.7 dB, which is worse than the simulated datasets above. We predicted the latent z𝑧zitalic_z for the full dataset, then sampled three points along the first component axis and generated the corresponding volumes	ABC	BAC	CAB	ABC	Selection 3
A: ML, VG, MS, SJR were supported by SNSF grants CRSK-3_190526 and 310030_204938 awarded to SJR. EB: Upschulte and TC: Dickscheid received funding from Priority Program 2041 (SPP 2041) “Computational Connectomics” of the German Research Foundation (DFG), and the Helmholtz Association’s Initiative and Networking Fund through the Helmholtz International BigBrain Analytics and Learning Laboratory (HIBALL) under the Helmholtz International Lab grant agreement InterLabs-0015. The authors gratefully acknowledge the computing time granted through JARA on the supercomputer JURECA at Forschungszentrum Jülich.	ACB	BAC	ABC	CBA	Selection 3
A: A careful examination of these explicit terms may be of interest in future work, but we collapse them into single functions and continue these calculations with the aid of a symbolic package up to some chosen order k𝑘kitalic_k: B: Three-body interactions are apparent, as are the existence of interactions between coupling and heterogeneityC: These terms, with b=0𝑏0b=0italic_b = 0, are the second-order interaction terms from [69]	ACB	BAC	CAB	ACB	Selection 3
A: In Sec. III, for the case of noninteger α𝛼\alphaitalic_α, we construct a first-order perturbation theory for the effect of noise on the characteristic function and derive macroscopic observables: population-mean voltage and firing rateB: In Sec. IV, the theoretical results for macroscopic states of homogeneous populations of QIFs are reported. In Sec. V, we compare the theoretical results against the background of the results of numerical simulation (also for a less mathematically challenging case of heterogeneous populations) and discuss general implications of the theoretical results and the examination (Appendix C) of the possibility to construct the generalization of the pseudocumulant expansion to the case of noninteger α𝛼\alphaitalic_αC: In Sec. VI, conclusions are drawn.	CBA	ABC	BCA	BAC	Selection 2
A: Predictive coding can be performed over any sensory modality that has some temporal sequenceB: As natural speech forms a cognitive map, predictive coding may underlie the geometry of human languageC: Intriguingly, large language models train on causal word prediction, a form of predictive coding, build internal maps that support generalized reasoning, answer questions, and mimic other forms of higher order	BAC	ABC	ACB	CBA	Selection 2
A: Temporal-spatial convolution is used with spatial modules, made with self and graph attention, to reveal spatial features of brain activityB: The linear layer is used to project the feature dimension.C: Then, the model obtains results by matching test data to templates. (B) Architecture of the EEG encoder	CBA	ACB	BCA	BAC	Selection 3
A: Even with these simplifying assumptions, species tree estimation is confounded by gene tree heterogeneity. Heterogeneity is particular problematic for concatenation-based methods, as the species tree for the entire concatenated sequence can disagree with gene trees for particular loci, Roch and Steel (2015)B: Many theoretical results, positive and negative, have been established when the only source of heterogeneity is ILS, see Degnan and Rosenberg (2006), Allman et al. (2011), and Mirarab et al. (2014). Incomplete lineage sorting is modeled by the multispecies coalescent (MSC) model.C: Common sources of heterogeneity include incomplete lineage sorting (ILS) Rannala and Yang (2003), horizontal gene transfer (HGT) Roch and Snir (2013), and gene duplication and loss (GDL) Arvestad et al. (2009)	BCA	ACB	ABC	CBA	Selection 2
A: In fact, Lawley and Madrid [24] have extended these works to also consider the effect of small noiseB: Relevant to our paper, they argue that the relative benefit of increased numbers of walkers is relatively weak (scaling as 1/log⁡(N)1𝑁1/\log(N)1 / roman_log ( italic_N )), and so it remains unclear if many biological systems are in this regimeC: We also note the work of Weiss [42], who determined first-hitting-time estimates when the walkers are initially uniformly distributed throughout the domain.	BAC	ACB	ACB	ABC	Selection 4
A: A recent work demonstrated that the weight uncertainty with the form of SaS structure can be also incorporated into the transformer [45]B: In addition, gated recurrent neural networks with multiplicative mechanisms were recently shown to be able to learn to implement linear self-attention [46]C: Furthermore, the relationship between linear transformers allowing for faster autoregressive learning and RNNs was established in a recent work [47]. Taken together, our current work would be a starting point to establish the bridge between the biological learning (towards the science of specialized brain circuits) and transformer learning within the seminal predictive coding hypothesis, which can be put in the theoretically solid variational free energy minimization conceptual framework.	CAB	BCA	BAC	ABC	Selection 4
A: Data acquisitionB: The images were captured using Leica DMi8 microscope (Leica) equipped with 10×/0.32 objective lens. We obtained one whole slide image from each group.C: Bright-field images used in this paper were obtained under the protocol described in [5]	ABC	BAC	CAB	ACB	Selection 4
A: In simple terms, an ANN learns from data by adjusting the weights of the connections between nodes in order to minimize a loss function that measures the difference between the desired output and the actual output of the network. More specifically, the optimization step is performed using the Stochastic Gradient Descent (SGD) algorithm, which iteratively updates the weights of the network by moving them in the direction of the steepest descent of the empirical loss function of a single training sampleB: Such a mechanism does not seem to be biologically plausible for BNNs, as many authors have pointed out. Parameter update in BNNs occurs only locally, and distant neurons are only indirectly connected through the endogenous reward system. This observation is closely related to the weight transportation problem [6, 2, 4]. We refer to [12, 11] for a detailed discussion about the role of SGD in BNN, which the author of [10, Section 5] summarizes as follows:C: The gradient itself is computed with the so-called backpropagation algorithm. In particular, the update of any parameter is based on the states of all other parameters	ACB	BAC	ABC	BAC	Selection 1
A: The formation of NCNM was formed by a self-assembly of homogeneous nanoparticles with negative surface charge in the desired shallow channel using Laplace pressure to halt the solvent at the base and evaporation of solvent. A close-packed fcc was formed by the growth of the ordered lattice induced by the evaporation.B: The fabrication of microchannel and formation of the NCNM for the fluidic memristor is similar to previously reported methods [36, 37] and is described in the SI in detail. A master for multi-layered channels (target heights are 5 μ𝜇\muitalic_μm for shallow channel and 100 μ𝜇\muitalic_μm for deep) was created using a multi-step UV exposure with negative photoresist (PR, SU-8 2005, 3050, Microchem Co., USA)C: After surface treatment of the master with (3,3,3-trifluoropropyl)silane (452807, Sigma-Aldrich, USA) for easy separation, Polydimethylsiloxane (PDMS, Sylgard, Dow Corning Korea Ltd., Korea) was poured and cured by heating. The detached PDMS device was bonded with a slide glass	CBA	CAB	ABC	ACB	Selection 2
A: It offers an extensive description of the trade-off between disease control and intervention costs applied in an infectious disease model and can be additive. Based on the considerations mentioned earlier, the following OCP is formulated by a hybrid cost function combining linear and quadratic terms, where the controls are shown in red for emphasis.B: Different types of cost functions may be appropriate for different scenarios, such as an exponential cost function avusuglo2023leveraging ; grandits2019optimal , where it can be used if the decision-maker under uncertainty, and its concave nature of exponential cost describes a risk-averse attitude, or a quadratic cost function, which its convexity ensures that any local minimum is also a global minimumC: Besides, the choice of the cost function is contingent upon the specific goals of the optimization problem	BCA	CAB	ACB	CBA	Selection 4
A: They control voluntary movement and also make a crucial role in cognitive processes (e.g., action selection)B: Dysfunction in the BG is related to movement disorder (e.g., PD) and cognitive disorder.C: The BG exhibit diverse functions for motor and cognition	BCA	ABC	ACB	CBA	Selection 1
A: Therefore, future work should aim to compute and store the Rashomon set of a wider variety of model classes. Future work may investigate incorporating Rashomon sets that may be well-approximated (e.g., GAMs, [10]), but not computed exactly, into the RID approach. Nonetheless, sparse trees are highly flexible, and using them with RID improves the trustworthiness and transparency of variable importance measures, enabling researchers to uncover important, reproducible relationships about complex processes without being misled by the Rashomon effect. B: RID can be directly computed for any model class for which the Rashomon set can be found – at the time of publishing, decision trees, linear models, and GLMsC: A limitation is that currently, there are relatively few model classes for which the Rashomon set can be computed	CAB	BAC	ABC	ABC	Selection 1
A: We report our performance against five leading methods [24, 41, 15, 32, 37]B: We detail our experimental methodology in Sec. 5.1 and report quantitative and qualitative evaluations in Sec. 5.2. Ablation studies are presented in Sec. 6. C: Following common practice in the field, we evaluate DREAM with the largest public neuroimaging dataset, the Natural Scene Dataset [2]	CBA	BCA	BAC	ABC	Selection 2
A: We note that the optogenetic treatment could have benefits in comparison to the traditional deep-brain-stimulation (DBS) treatmentB: The DBS has the following disadvantages OG3 ; OG4 ; (a) it is difficult to accurately determine the target cells, leading to cause many side effects and (b) a process with many trial and errors is necessary to each patient for optimal controlC: On the other hand, the target cells can be accurately located by optogenetic stimulation. Hence, side effects and trial-and-error process may be decreased in spite of limitations for application to the human patients OG3 .	CAB	CAB	CBA	ABC	Selection 4
A: Bürger and Lynch, 1995; Gomulkiewicz and Holt, 1995; Matuszewski et al., 2015). For models like this, our findings may be of relevance either for the initial response (in case of a sudden shift) or even in the long term (in case of a moving optimum). Below, we briefly review recent treatments focusing on the polygenic dynamics underlying trait evolution. B: Because we assume exponential directional selection, beneficial mutations, and weak random genetic drift, all mutants that become ‘established’ will sweep to fixation in the long run. Therefore, we focus on the initial response when characterizing patterns of adaptation as sweep-like vs. shift-likeC: In Section 6.4, we argue that our results are of relevance for other modes of directional selection. A popular model is one in which a population is initially in equilibrium under mutation-stabilizing selection-drift balance and then starts to evolve due to a sudden or a continuous change in the environment, modeled by a sudden shift in the optimum phenotype or by a continuously moving optimum (e.g	CAB	ACB	CBA	BAC	Selection 1
A: In this study, we aim to enhance GNNs to better capture long-range interactionsB: This novel approach reduces interaction distances between nodes to a single hop. Extensive experiments on four long-range graph benchmarks validate our method’s ability to enhance any GNN to capture long-range interactions. C: We achieve this by transforming original atoms into neural atoms, facilitating information exchange, and then projecting the improved information back to atomic representations	BAC	BCA	ACB	BAC	Selection 3
A: In one set of experiments, in order to study dependence on model size, we take a selection of LLMs from the Pythia family [22]B: We provide full details of the experiments in the Supplementary Information. Throughout the paper we primarily investigate the open source GPT-J model with around 6 billion parameters [5]C: The code for all experiments is provided at github.com/rmldj/memory-llm-paper.	ACB	ACB	BAC	ABC	Selection 3
A: Our analysis has focused only on reconstructing the covariance between the taxa in a set of samplesB: To understand more general relationships, including across kingdoms, other methods should be employed alongside covariance network reconstructionC: In particular, non-negative tensor factorization methods show promise in identifying relationships between groups of taxa. Two popular tools, MOFA+[37] and DIABLO[38] are both designed to use non-negative tensor factorization to identify groups of related variables across data types.	CAB	BCA	ABC	BCA	Selection 3
A: The virus is now managed similarly to seasonal influenza, with new vaccine booster shots tailored to target emerging virus variants. B: At present, daily life in most countries has returned to pre-pandemic normsC: The development and widespread use of vaccines and treatments eventually brought an end to the pandemic, officially declared by the World Health Organization (WHO) on May 5, 2023[1]	CBA	BCA	ACB	ABC	Selection 1
A: Since each p𝑝pitalic_p-dimensional simplex corresponds to a collection of (p+1)𝑝1(p+1)( italic_p + 1 ) data points222Recall that points are 00-dimensional simplices, edges are 1111-dimensional, etc., we can assign weights to the datapoints themselves by considering for each node in the Vietoris-Rips complex, the sum of the weights of its cofacesB: For example, given a 1111-dimensional harmonic representative, we can translate the weights from the edges to the nodes by assigning to each node the sum of the weights of its adjacent edgesC: This give us a set of weights on the nodes for each bar.	CAB	BCA	CAB	ABC	Selection 4
A: In iopart.cls, to make any line start at the left margin of the page, add \fl at start of the line (to indicate full left).B: The iopart.cls class file automatically does this and indents each line of a displayC: The normal style for aligning displayed equations in our published journal articles is to align them left rather than centre	ACB	BAC	CBA	ACB	Selection 3
A: As evidenced in FigB: However, the pre-trained LMs are not inherently trained on data specific to the target environmental ecosystems, which often results in a failure to effectively capture feature dependencies within the descriptionsC: 3 (b), the input features from samples in different seasons tend to be mixed in the latent embedding space of the original pre-trained DistilBERT model. This indicates that a standard LM, when applied directly, may fall short of capturing semantics from the text generated for our target task.	CAB	CAB	BAC	ACB	Selection 3
A: Nucleotide Transformer introduced the first large-scale transformer-based DNA LMs. All NT models share the same architecture, but differ in their number of training genomes and model parametersB: A second generation of multispecies models released later (NT-V2) extended the input length to 12,282 bp. The NT models were evaluated on tasks comprising promoter, SS, histone modification and enhancer prediction with a context length of up to 600 bp. C: Models were trained on either the human reference genome, 3,202 different genetically diverse human genomes or a selection of 850 genomes from a range of species. To increase the receptive field of the model, sequences were tokenized as 6-mers, allowing for processing sequences of up to 5,994 bp in length	ABC	ABC	ACB	BAC	Selection 3
A: However, from a statistical perspective, concerns arise regarding the potential for false discoveries of genes that lack genuine spatial variability. This concern becomes more pronounced when a large number of genes are simultaneously tested across most frameworks. If the false discovery rate or type 1 error is not adequately controlled, it may lead to incorrect conclusions and the selection of numerous genes that exhibit false spatial variability.B: The mathematical models employed for capturing the data generation process and the innovative model-free SVG detection technique have proven valuable for uncovering significant SVGs that offer critical biological insightsC: We have previously discussed both model-based and model-free methods for detecting SVGs	BAC	BCA	BCA	CBA	Selection 4
A: As shown in Figure 2, MolIG, which utilizes both pre-training and data augmentation strategies (represented by the grey bar), performs best among all model architecturesB: Models without pre-training (represented by ’w/o pretrain’) perform worst in all tasks. Pre-training strategies without data augmentation (represented by ’w/o img aug’) perform second best, yet they show a significant performance increase compared to models without any pre-training. Excluding either of these two components can easily lead to a decrease in performance.C: This ablation study aims to evaluate whether pre-training strategies and data augmentation strategies can help the model achieve better performance in downstream tasks	ACB	BCA	CAB	ABC	Selection 2
A: During the downstream task, the classification head is attached to the pretrained encoder and its weights are fine-tuned on the dataset of said task in a fully-supervised manner. The architecture of the ”downstream model” is used as a baseline in all of our experiments, where all of its layer weights are adjusted during fully-supervised training. B: The weights of the encoder or ”backbone network” are frozen after the pretraining phaseC: Figure 2: The architecture of the CNN encoder trained via SSL contrastive learning (left) and the classification head trained for the downstream task (right)	CBA	ACB	BAC	ABC	Selection 1
A: Our main analysis encompasses two distinct definitions of tumor recurrence: a broad definition including edema, enhancing core, and necrotic core, and a more specific definition aligned with current RANO guidelines focusing only on the enhancing coreB: Furthermore, we conduct a patient-specific comparison with the Standard Plan, classifying a model as ’Better’ if it provides higher coverage for an individual patient and ’Worse’ if it falls short of the Standard Plan’s coverage. In instances of equal coverage (e.g., both achieving 100%), we label the outcome as ’Equal’. The comparison outcomes, including average results, are depicted in Figures 5c and 6c. C: Figures 5 and 6 present the mean and standard deviation of Recurrence Coverage for these definitions	BCA	CAB	ACB	CAB	Selection 3
A: Figure 2: Time course of C𝐶Citalic_C derived by solving Eqs. (3)–(4) for the same parameter values as in Fig. 1B: In (a) and (b) we show the time evolution for two initial conditions to illustrate that the only fixed point of the system is stable but excitableC: In (c) there is a stable limit cycle.	ABC	BAC	BCA	BAC	Selection 1
A: The top three values for each metric and each track are marked as gold, silver and bronze cells in decreasing order. A ‘’ behind the team name means that the segmentation predictions have been converted from the multiclass submissions and inserted here. If a team only submitted to one of the tracks the columns of the other track are filled with a ‘-’.B: Table 4: Binary segmentation task results (mean ±plus-or-minus\pm± standard deviation) for MRA and CTA in Dice scores, centerline Dice (clDice) scores and errors in the zero-th Betti number β0subscript𝛽0\beta_{0}italic_β start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPTC*: The arrow indicates the favorable direction	CAB	BAC	BAC	ACB	Selection 1
A: Building upon DDPMs, MoFusion (Dabral et al. 2023) employs U-Net (Ronneberger, Fischer, and Brox 2015) as the backbone for the denosing kernel in motion sequence synthesis. Apart from applications on continuous data, many research efforts are devoted to discete data generationB: EDP-GNN (Niu et al. 2020) introduce Score SDEs, a learning paradigm of diffusion models, to the generation of discrete graph adjancency matrices.C: The formation of human motion skeletons share similarities to molecules, as both are represented by point clouds connected by edges. The difference is that human motion generation does not require predictions of edges	ABC	ACB	BAC	BCA	Selection 4
A: In this diagram, every point is represented by a cell enclosed by the red solid lines and dashed curves, then the surface area can be calculated with such information.B: The purpose of this step is to obtain the dual complex 𝒞𝒞\mathcal{C}caligraphic_C and the conjugated Laguerre diagramC: As shown in Figure 1, the black points are the center of atoms and dashed lines connecting black points denote the edges in the tetrahedrization	BAC	BCA	CBA	BAC	Selection 3
A: 2021) and ChemBERTa-2 (Ahmad et al. 2022). B: Additionally, in our experiments, we tested two types of M-Encoder: CHEM-BERT (Kim et alC: Since any molecular pretraining model can serve as the M-Encoder in our method, in this section we mainly describe the Transformer encoder block (TEB), the fundamental component composing mainstream language models, as well as the internal structure of the MT-Encoder	ACB	CBA	BCA	ACB	Selection 2
A: Of course, this qualitative understanding is no substitute for the detailed analysis obtained by solving the replicator equation and, in particular, cannot predict the existence of an all-cooperators regime. B: Direct application of the variance decomposition formula Weiss_2005 shows that a non-degenerate distribution of group sizes always increases the variance of a cooperator’s fitnessC: A fundamental result of Wilson’s approach is that cooperation is promoted by increasing the variability of group composition or, if fitness is not a linear function of the number of cooperators, by increasing the variance of a cooperator’s fitness	CAB	CAB	CBA	ACB	Selection 3
A: In exploring the influence of racial and ethnic groups on BP values, we categorized the final pre-processed dataset based on the racial/ethnic groups listed in Table IIB: Importantly, race and ethnicity are compound factors, impacted by biology (genetic factors that impact BP from a physiological or anatomical standpoint) and also by social determinants of health (factors related to race/ethnicity disparities)C: Table II presents the mean and STD for DBP and SBP across racial and ethnic groups. Accordingly, the African American or Black group had the highest mean SBP and DBP values. The Asian population demonstrates the lowest SBP, while the Hispanic group has the minimum DBP.	BAC	BCA	ACB	ABC	Selection 4
A: The effects of repressing the clock can be seen in Figure (4A) which shows the dynamics of the core circadian oscillator gene BMAL1. The failure of BMAL1 to oscillate indicates the repression of the circadian clock. Figure (4B) shows the filter output for BMAL1. B: In order to demonstrate the scalability of our approach on real world datasets, we used the PKF to analyze time-course gene expression measurements (RNAseq) of a circadian clock dataset in human cancer cell lines [rebekah_clock] with contains of roughly 1.8 million gene expression measurements consisting of 32337 gene expression measurements across two conditions and 14 time-points with two replicates per time-point and condition. Briefly, the mammalian circadian clock is an cell-autonomous oscillatory genetic circuit which regulates diverse cellular functions by synchronizing them to daylight patterns [takahashi], [sgolden]C: For a brief description of the mammalian clock, refer to Appendix A.5.2. In this dataset, the circadian clock is disrupted by activating the gene MyC [rebekah_clock] which causes the clock to be repressed	CAB	CBA	CBA	CBA	Selection 1
A: We propose Absorb-Escape, a generalisable post-training algorithm for refining the quality of generated discrete sequencesB: We show that Absorb-Escape further increases the performance of DiscDiff by 4% in long DNA generationC: In addition, Absorb-Escape allows control over the property of generated samples.	ACB	CAB	ABC	BAC	Selection 3
A: FM-indexes over minimizer digests are known to be usually significantly faster than indexes over the original datasets, both because some characters are not represented in the digests and because we use a backward step for each minimizer rather than for each character, incurring fewer cache missesB: If we are willing to sacrifice the true-positive rate moderately, we can increase w𝑤witalic_w and also achieve significant speedupsC: Interestingly, however, we achieved slightly better speedups with both kernelization and minimizers than with either separately.	ABC	BAC	CBA	ABC	Selection 2
A: We base our kernels on the vLLM implementation ([51]).B: Fused kernels enable complex operations, such as Rotary Embedding ([50]) which would otherwise be executed sequentially, to be combined into a single kernel than can be executed in parallel on a GPUC: To optimize the forward pass, we implement fused CUDA kernels	CBA	BCA	BCA	ACB	Selection 1
A: This feature denotes the level of light scatter at a 90-degree angle in relation to the laser beamB: Side Scatter (SSC)-Cell’s granularityC: SSC reflects the internal complexity or granularity of the cell, encompassing features like the presence of granules, nuclei, or other organelles. A high importance of this feature may suggest that either our model can differentiate between cells that have different degrees of complexity, such as lymphocytes, monocytes, and granulocytes, or our model can identify cells that have abnormal granularity, such as blast cells or malignant cells.	BAC	ABC	ACB	CAB	Selection 1
A: The natural deathB: both phases of the pandemicC: The recovery periods σ−1superscript𝜎1\sigma^{-1}italic_σ start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT and γ−1superscript𝛾1\gamma^{-1}italic_γ start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT are taken to be 21⁢days21days21\,{\rm days}21 roman_days	CBA	BAC	CAB	BCA	Selection 3
A: Parameter reallocation within the network also plays an important role in memory conservationB: This step involves redistributing parameters by expanding channel width in critical modules while reducing it in less important onesC: This strategy allows for more efficient learning of representations in higher-dimensional spaces without losing feature information, while also speeding up inference by removing redundant parameters.	CAB	BCA	ABC	CAB	Selection 3
A: Any approximation to 𝐊𝐊\mathbf{K}bold_K that accounts for identical repetitions, including all known lossless statistical compression algorithms, can achieve equivalent or superior results, as demonstrated in the Supplementary Information and [4]B: This alignment with ‘Template Program A’ effectively highlights the association of AT with well-established principles of compression and coding theory, thereby refuting the initial claim of its authors to present a unique methodologyC: Moreover, the authors’ suggestion that their index may be generalised as a universally applicable algorithm for any object (including text) [31] further underscores the	CAB	ABC	CBA	BAC	Selection 2
A: This tactic for Cognitive Security against readout requires alteration, albeit strategically minimal alteration. And as seen with the similarities between SMON/SION and MOA/IOA, there is a meaningful analogy between AA and defensive measures against RA, and between RA and defensive measures against AA. If no readout is made of cognition and neural activity, then alteration may not in general be reliably detectedB: Finally, there is an interesting notion which emerges from the discussion of attack and defense methods. The Bonaci et al BCI Anonymizer is an innovative approach to the proper design of BCI hardware, but remains vulnerable to air-gap attacks, malicious implants, or BCI hardware vulnerabilities. In order to maximally mitigate such vulnerabilities, one would likely need to induce physiologic noiseC: The roles of the attacker and the defender exist as odd sorts of mirror images in both cases. While we do not currently see any path towards verifying this one way or another, it leads us to make the following conjecture. There may be no means to achieve physiology-level security guarantees against readout without enacting some degree of alteration to cognition itself, nor such guarantees against alteration without performing some readout. As with much of the myriad implications of this paper and the problems it seeks to begin addressing, there is a great deal of future work to be done on the topic.	CAB	BAC	BCA	BCA	Selection 2
A: These metrics are evaluated both within a batch of 64 samples and across the entire test set, following the methodology outlined in Liu et al. (2023b). As presented in Table 1, MolBind’s performance surpasses that of MolCA by a large margin. Overall, MolBind significantly outperforms prior work on two molecule-language retrieval tasks, validating its effectiveness in aligning the molecule and language modalities. B: We evaluate the graph-language and conformation-language retrieval performance on the MoBind-M4 dataset. For our baselines, we employ Sci-BERT Beltagy et al. (2019), KV-PLM Zeng et alC: (2022), MoMu Su et al. (2022), and MolCA Liu et al. (2023b), assessing their performance using Recall@1 and Recall@20	CAB	ACB	BAC	BAC	Selection 1
A: If more than one route is found for a molecule, the route scored the highest by AiZynthFinder is retained. For solved routes, we compute the depth of the tree. Since the maximum tree depth allowed by AiZynthFinder’s default parameters is 7, we assign a depth of 10 for molecules where a route is not found. B: We use AiZynthFinder [11] with default parameters to compute the routes of both datasetsC: A route is considered to be solved if all leaves are purchasable molecules	ACB	CAB	ABC	ABC	Selection 2
A: The three panels summarize the system behavior for α=0.8,0.4𝛼0.80.4\alpha=0.8,0.4italic_α = 0.8 , 0.4, and 0.2, where the inner invasion in odd-labeled loop is strong, intermediate or weakB: Our first results, shown in Fig. 3, were obtained when the interaction between the triplets are strongC: Evidently, the probability α+δ𝛼𝛿\alpha+\deltaitalic_α + italic_δ cannot exceed 1, which explains the different maximal δ𝛿\deltaitalic_δ values for each panels.	BCA	BAC	BCA	ACB	Selection 2
A: We compare several transformer architectures on the task of cancer detection on a single ROIB: We choose a modified ResNet18 [9] as our ROI-scale baseline, with only one sequence of convolutions and batch normalization in each residual block. This reduction in the number of parameters mitigates overfitting and is associated with an increase in performance[2]. C: To that end, we use a standard Vision Transformer[6] (ViT) architecture, a Compact Convolutional Transformer[7] (CCT), and a Pyramid Vision Transformer[8] (PvT)	BCA	CAB	ACB	CBA	Selection 3
A: One response was removed because the person stated in the open ended comments that they did not use the tool, but filled out the survey anyway. User responses indicated that the dashboard were visually appealing (>89% appealing or very appealing) sufficiently easy to navigate (>82% responded easy or very easy), provided a positive user experience (>91% satisfied or very satisfied), and trusted the accuracy of the data (>86% trust or strongly trust). B: Surveys were sent to 78 users of the Grafana v3 dashboard to evaluate experiences with a 35.89% response rate (n=28)C: The GEMS Sensing user interface evolved from simple HTML and CSS pages (Figure 4A) into a REACT.JS frontend (Figure 4D) and then into a Grafana-based set of organizations and dashboards (Figure 4G)	BCA	ABC	CBA	BAC	Selection 3
A: These MLMs are available in the HuggingFace transformers library Wolf et al. (2020). However, it can be noted that some level of adaptation was needed at the tokenization step to use some of the models within NLstruct, especially for frALBERT and FlauBERT models555The source code is available here: https://gitlab.lisn.upsaclay.fr/nlp/deep-learning/nlstruct. B: To compare French MLMs, we train one NER model using each of the language models in our benchmarkC: The text encoder component in NLstruct relies on embeddings produced by a BERT language model, a char-CNN encoder, and static French FastText embeddings	BAC	ABC	CBA	CAB	Selection 3
A: The proposed method has a wide variety of potential applications, including drug discovery, and molecular designB: By creating low-dimensional embeddings and using them to find molecules with related qualities, the approach makes it possible to construct unique compounds with certain properties. Overall, this signifies a promising avenue for molecular structure analysis, employing kernel methods to unlock new possibilities. Following are our contributions:C: It offers the opportunity to quickly search through vast datasets of chemical structures in search of compounds with desirable properties	BCA	ABC	CBA	ACB	Selection 4
A: Specifically, we first introduce indices evaluating signal distortion by the gain and the phase delay characteristics and derive these characteristics of MC channels based on the diffusion equation and the rate equationB: We then show design conditions for MC channels in which the magnitude of distortion becomes below a specified level based on the indices. Using the proposed method, we demonstrate the design procedure of specific MC channels that satisfy given specifications. Finally, the roles of MC channels in nature are discussed from the perspective of signal distortion.C: In this paper, we propose a method to analyze signal distortion caused by one-dimensional diffusion-based MC channels	CBA	BAC	BCA	CAB	Selection 3
A: The prediction-based method, a learning-based approach, employs a deep neural network to extract meaningful features from past statesB: It leverages the neural network’s ability to automatically learn and remember states without requiring manually designed storage solutions and also promotes more efficient explorationC: However, this approach cannot effectively encourage exploration as RL training progresses, particularly when the agent needs to escape out of local optima. Specifically, when the neural network exhibits unexpectedly high generalization performance, it may predict low errors even for unvisited states and hinder the agent from exploring the space further.	ACB	CAB	ABC	BAC	Selection 3
A: 2023b; Zhang et al. 2022b)B: Recently, a growing body of literature has explored the integration of SNNs with DRL (Zhang et al. 2022a; Wang et alC: Several methods convert a trained DQN into a SNN version (Patel et al. 2019; Tan, Patel, and Kozma 2021) or directly train a deep spiking Q-network (Liu et al. 2021; Chen et al. 2022). These endeavors yield competitive performance on Atari video games with discrete action spaces compared to the original DQN.	BAC	BCA	CAB	ACB	Selection 1
A: We remove the data augmentation during the second training stage, which means training solely using the DFS order that can generate canonical SMILESB: Data AugmentationC: Table 6 demonstrates a significant decline in model performance across all metrics. This clearly demonstrates that our data augmentation significantly improves the model’s performance.	BCA	BAC	ACB	ACB	Selection 2
A: For flow-based models, despite exhibiting high uniqueness and novelty, their validity was lower, specifically less than 89.03%percent89.0389.03\%89.03 %, which significantly trails behind InstGAN. InstGAN demonstrated comparable performance to SOTA diffusion models GDSS and D2L-OMP. All models, including pretraining and tasks involving single properties (QED, logP, SA), as well as multi-property tasks (QED, logP, SA), achieved an overall score surpassing 93.87%. Among GAN-based models, while MolGAN and TransORGAN exhibited a novelty of 100.0%, their validity and uniqueness fell significantly lower than InstGAN. Specifically, MolGAN demonstrated a mere 4.3% uniqueness, and TransORGAN exhibited a validity of only 74.31%. ORGAN’s validity stood at 67.96%, markedly lower than InstGAN’s impressively high validity exceeding 95.45%B: Furthermore, InstGAN outperformed SpotGAN in terms of validity, uniqueness, novelty, and total score, with the highest overall performance. InstGAN was trained for single- and multi-property optimization. In single-property optimization, InstGAN was trained individually using QED, logP, and SA. In multi-property optimization, these three properties were used to jointly train InstGAN. The validity, uniqueness, novelty, and total score all reached up to 93.87%percent93.8793.87\%93.87 %. Overall, while InstGAN generated molecules from less informative SMILES strings compared to graph-based models, it surpassed VAE-, flow-, and GAN-based baselines and demonstrated comparable performance to SOTA diffusion models. This showcases InstGAN’s robust capability in molecular generation, excelling in both single- and multi-property optimization. C: Table 2 compared the results of InstGAN with various baseline models (including VAE-, flow-, diffusion-, and GAN-based models) for chemical property optimization on the ZINC dataset. To ensure a fair comparison, InstGAN was pretrained several times, and the average results are presented. Additional details on these multiple pretraining sessions are provided in Appendix C. For VAE-based models, although RNN-Attention and TransVAE generated molecules that were highly unique and novel, their validity (i.e., <71.6%absentpercent71.6<71.6\%< 71.6 %) was significantly lower compared to InstGAN. InstGAN outperformed Character-VAE and Grammar-VAE in all metrics. Although JT-VAE exhibited high validity and novelty, the uniqueness was only 19.75%percent19.7519.75\%19.75 %, significantly lower than that of InstGAN	CBA	CAB	CAB	BCA	Selection 4
A: It may be that one or a few self-supervised tasks are insufficient to learn effective representations, as mentioned in [22]B: To further showcase the effectiveness of our method, we compare our algorithm with some recent protein pretraining language models on fold classification [68, 20, 37, 85, 94], which are typically pre-trained with a large scale of data. As shown in Table 6, our method still yields better results without any pre-training or self-supervised learningC: This sheds light on the direction of our future efforts: it is interesting to embrace our algorithm with existing protein language models since our core idea is principled.	BCA	BCA	BAC	ABC	Selection 3
A: Advancing in these two directions constitutes the central motivation of the present work.B: The systematic investigation of whether each type of heterogeneity enhances or diminishes the probability of infection has not been systematically undertaken in previous studiesC: Furthermore, a link between heterogeneous infectivity and microbial growth mechanisms (e.g., microbial rates of birth and death) has yet to be established	BAC	BAC	BCA	CAB	Selection 4
A: Target novelty (N): The percentage of novel molecules among all the desired valid molecules, the novel molecule is different from training samples;B: 3C: Target validity (V): The percentage of valid molecules among all the desired molecules, which is measured by RDkit (Landrum et al., 2016) and widely used for calculating validity (Hoogeboom et al., 2022; Xu et al., 2023)); 4	CBA	ABC	CAB	ABC	Selection 3
A: Each training image is shown 4 times, and each test image is shown 80 times. We took the average of all trials for each image to form the final dataset. B: Each trial in the data is from -0.2 seconds to 0.8 seconds relative to the onset of the stimulusC: The training images and test images are presented in separate sessions, but within training and test images all the orders are pseudo-randomized	ABC	CAB	BAC	ABC	Selection 2
A: Our drug-ASE matrix analysis (FigB: 6) to data from Google Trends (Fig. 7), we found that the ASEs mentioned most frequently over time are nausea, pain, vomiting, and constipation.C: 4) highlights numerous prevalent ASEs based exclusively on social media reports. Additionally, comparing mentions of ASEs on social media in intervals of 14 days (Fig	BCA	ACB	ABC	BCA	Selection 2
A: Our approach retrieves drug information using a set of free-to-use application program interfaces (APIs) provided by NLM via representational state transfer (REST) to access the RxNorm and RxTerms datasets: the RxNorm and RxTerms [9], and RxClass [10] APIsB: To avoid sending potentially thousands of requests to the NLM servers, we also make use of RxNav-in-a-Box, a locally installable version of the APIs, in combination with Docker [11]C: To query the APIs, we created an R [12] package interface which is available for download on GitHub.	BAC	BAC	BCA	ABC	Selection 4
A: This modification removes sampling error in the fragment generation process, creating an artificially easier learning problem that should result in better performance. We emphasize that ICEBERG-ADV is only suitable for benchmarking: in most realistic C2MS problems ground truth spectra are not available at inference time, meaning that MAGMa cannot be applied.B: The stages are trained independently, with the first stage approximating the output of a heuristic fragmenter, MAGMa (Ridder et al., 2014), that uses signals from the spectrum to find a minimal set of fragments that explain the peaks. We introduce ICEBERG-ADV (advantaged), a variant of ICEBERG that replaces the first stage of the model with the exact MAGMa outputC: ICEBERG (Goldman et al., 2024) is a structured method for high-resolution spectrum prediction that uses a two-stage model. The first stage autoregressively predicts a set of fragments, and the second stage predicts a mapping between those fragments and the peaks in the spectrum (see Appendix C)	CAB	CBA	BAC	ACB	Selection 2
A: In addition, we do not cover web-based tools, software and libraries that implement prediction methods (see e.g. [8] for a review). B: We limit our focus to new methods that have been tested against large benchmark datasets on pre-clinical synergy and dose response data, in particular leaving out drug-drug interaction (DDI) prediction, which has its own focused literature (see e.gC: [7]), as well as papers that are using small scale data or lacking comparison to alternative prediction methods	BCA	ACB	BCA	CAB	Selection 4
A: The complexity K⁢(S)𝐾𝑆K(S)italic_K ( italic_S ) measures the number of distinct transformations, including their execution conditions, allowed by the graph transformation modelB: Such a complexity measure is useful for study and evaluation of empirically inferred input transition systems. In the case of chemical reaction networks, for instance, the complexity allows for formal reasoning about the underlying model of chemistry, opening up possibilities for comparison and further analysis of various industrial and biological processes. C: The measure K⁢(S)𝐾𝑆K(S)italic_K ( italic_S ) thus captures the amount of elementary behaviors the model exhibits, or in other words, the diversity of the model dynamics	ACB	BCA	BCA	ABC	Selection 1
A: These works hypothesise a minimal thermodynamic cost or entropy production per product made, which is related to the information accurately transferred from template to products [2, 31, 32]B: However, entropy production in stochastic thermodynamics is related to the relative rate of forwards and backwards transitions [33], rather than the relative rate of two different transitions; it therefore places no minimal cost on the specificity of a catalyst [29]. Exactly how these two ideas are reconciled is unclear. C: To the best of our knowledge, only Refs. [2, 31, 32] have considered the implications of two or more pathways for production and destruction of an ensemble of products, in each case without considering a detailed model of dynamical pathways	ACB	CAB	BCA	BAC	Selection 3
A: In addition to clustering, the software includes a feature for outlier detection to identify residues that deviate significantly from the expected torsion angle distributionsB: The software employs a distance-based approach to identify outliers, where residues that fall beyond a certain threshold euclidean distance from the cluster centroids are flagged as outliers with an ’x’ mark. Notably, empirical observations have shown that setting the threshold within the range of 85≤x≤10085𝑥10085\leq x\leq 10085 ≤ italic_x ≤ 100 yields results closely aligned with the validation reports provided by the Protein Data Bank (PDB) for the corresponding models. The outcome of introducing this into the program can be seen in figure 3.C: Outliers in the Ramachandran plot may indicate potential issues with the protein model, such as structural anomalies or misalignments	ABC	BCA	ABC	ACB	Selection 4
A: (2012). And individuals can be prone to stay away from those with similar opinions corresponding to out-group bias Kimura and Hayakawa (2008). Thus social relationships are under social biases.B: (2012); Fu et alC: Individuals can voluntarily stay away from those who do not adopt their opinions corresponding to in-group bias Holme and Newman (2006); Durrett et al	BCA	CAB	CBA	BCA	Selection 3
A: However, though ABC can potentially be expanded to high-dimensional applications, it is usually only practical to apply ABC when the parameter space is relatively low-dimensional. We propose to focus on inferences for the contagion parameters while considering the true network as a nuisance parameter that is marginalized over. This is most sensible if the purpose of estimation of disease parameters is to parameterize contagion models for the prediction of disease, especially if the model is meant to be generalizable over populations that exhibit varying contact structures but are affected by the same type of disease, such as in the study of outbreaks in two cities with differing population densities. B: The unknown contact network can be considered a latent parameter that can be jointly inferred on, alongside the contagion parameters of interestC: While MDN-ABC can readily be applied to epidemics on known networks \citepwang2023, it is often necessary to consider the uncertainty introduced by imperfect observations on networks	ACB	BAC	CBA	ABC	Selection 3
A: Here, we focused on the widespread standard consolidation theoryB: However, there exist other models and questions around systems memory consolidation conceptsC: For instance, the multiple trace theory suggests that some of the hippocampus patterns are conserved in the long term. This theory follows observations of hippocampal damages that produced temporally-graded retrograde amnesia only for semantic memories, but not for episodic ones (Nadel and Moscovitch, 1997).	BAC	BAC	ABC	BCA	Selection 3

**A**: The PCAM dataset was downloaded from the original website (https://github.com/basveeling/pcam). All images have a size of 96 x 96 pixels, in three colors**B**: The training set has 262,144 images (80 % of the total), the validation set has 32,768 images (10 %) and the test set also has 32,768 images (10 %). All datasets have a 50/50 balance between positive (tumor present) and negative (tumor absent) samples**C**: The patches of 96 x 96 pixels images were automatically extracted from the CAMELYON dataset [13]. For each image, a positive label indicates that the 32 x 32 pixel center of the image contains at least one pixel annotated as tumor tissue (Figure 1).

CAB

ABC

ACB

CAB

Selection 2

**A**: By definition, an evolutionary stable condition is surrounded by unfit traits, at least within an α𝛼\alphaitalic_α-radius**B**: This form of a fitness landscape is referred to as a fitness valley and has been studied in a special case in [8]**C**: Based on this, we introduce a measure for the stability of a coexistence equilibrium, connected to the width of the surrounding fitness valley.

BAC

ABC

CBA

BAC

Selection 2

**A**: We first compute the virus-free equilibrium and endemic equilibrium after establishing the basic reproductive number**B**: Subsequently, we extend our analysis to the stochastic counterpart, determining the basic reproductive number in the stochastic framework. This stochastic reproductive number becomes instrumental in discerning the dynamics of coronavirus extinction and persistence, contributing valuable information for disease control strategies.**C**: In this section, our objective is to derive sufficient conditions that elucidate the extinction and persistence of the disease, providing novel insights into the control of COVID-19

ABC

BAC

BCA

ABC

Selection 3

**A**: In Wang et al. (2017), persistent homology was shown to outperform topographic power maps. In (Yoo et al., 2017), center persistency was shown to outperform the network-based statistic and element-wise multiple corrections. In Chung et al. (2023b), persistent homology based clustering is shown to outperform k𝑘kitalic_k-means clustering and hierarchical clustering.**B**: In Lee et al. (2011, 2012), persistent homology was shown to outperform eight existing graph theory features, such as clustering coefficient, small-worldness, and modularity. Kuang et al. (2019) showed persistent homology-based measures can provide more significant group difference and better classification performance compared to standard graph-based measures that characterize small-world organization and modular structure**C**: In Chung et al. (2017b, 2019a), persistent homology was shown to outperform various matrix norm-based network distances. In Wang et al. (2018), persistent homology was shown to outperform the power spectral density and local variance methods

CBA

ACB

CAB

BAC

Selection 3

**A**: In this note, we propose a multiscale model for glioma development and spread which interconnects the dynamics of glioma cells, vasculature, and vascular endothelial growth factors. Glioma cells mainly spread according to the anisotropy of brain tissue or moving along the blood vessels therein [17, 38]. In order to provide nutrient and oxygen supply to its growing cell population, the tumor produces and releases in the extracellular environment growth factors, such as VEGFs, that attract endothelial cells**B**: The resulting angiogenesis boosts the growth of the neoplasm, leading to an abundantly vascularized and progressed tumor mass. Standard treatment usually comprises chemo- and radiotherapy, in a concurrent or an adjuvant manner. Thereby, both tumor and normal tissue are affected, although to a different degree**C**: A necrotic region is formed within the tumor mass, due to impaired tumor-associated vascularization. We also consider here the effect of an anti-angiogenic drug (e.g., bevacizumab [28]), aimed at reducing the affinity of VEGF receptors on ECs toward their ligand released in the extracellular space by the tumor cells. This therapeutic approach affects the EC tactic motility by reducing their bias towards (increasing gradients of) VEGFs and impairing their proliferation, which, in turn, leads to degradation of glioma through diminished nutrient supply. Starting from the subcellular level of interaction between cells, VEGFs, and tissue, we set up the corresponding kinetic transport equations (KTEs) for glioma cells and endothelial cells, and perform a (non-rigorous) parabolic limit to deduce the macroscopic system of reaction-advection-diffusion PDEs for the involved cell populations: glioma and endothelial cells. These equations are then coupled to the evolution of VEGF concentration, healthy tissue, and necrotic matter, which are considered directly on the macroscopic scale.

ABC

BCA

ACB

Selection 1

**A**: For comparison, two other types of brains are also included. Purple points represent lissencephalic brains, with hemisphere sizes measured as depicted in Fig. 3**B**: Red points denote quasi-gyrencephalic brains, measured as shown in Fig. 4 and Fig. 5. Data are collected from academic publications and www.brainmuseum.org. For species with multiple images, each image is measured individually. When possible, brain sizes are measured along the lateral axis, encompassing both hemispheres. **C**: Plot of gyral size versus brain size for gyrencephalic brains. Blue and green points indicate gyral sizes measured using different methods, as illustrated in Fig. 1 and Fig. 2, respectively

ABC

CAB

ACB

BCA

Selection 4

**A**: Often, novel drugs may not demonstrate efficacy in clinical trials despite thorough preclinical safety testing**B**: Detecting synergistic effects among approved drugs holds clinical significance, as drug repurposing can streamline the expensive and lengthy process of developing new drugs**C**: In the future, integrating in silico drug synergy discovery with ex vivo testing of selected anti-cancer drug combinations in patient-derived 3⁢D3D\mathrm{3D}3 roman_D organoid models could significantly enhance the clinical translatability of this approach.

CBA

ACB

BAC

Selection 4

**A**: 2F), there is an optimal number of chromosomes for evolving novel traits (Fig**B**: 2B).**C**: We found that the skewness of the distribution of genotypes coded in each chromosome is essential to determining the probability of evolving new traits. Because skewness peaked with a finite number of chromosomes (Fig

BCA

ABC

CBA

CAB

Selection 1

**A**: The gradient component sum to zero along any cycles**B**: The curl components are zero for edges that are not a 2-simplex boundary and the entries sum to zero around each node. The harmonic component sums to zero around each node, and it also sums to zero along each 2-simplex. We tested the topological equivalence of female brain networks and male brain networks using the Wasserstein distance (5).**C**: We further determined if we can detect topological differences in the decomposed components (Figure 2)

BCA

CAB

ABC

ACB

Selection 1

**A**: preprint arXiv:2210.05819).**B**: Others are modifications, such as models with varying population sizes [Mohle2002, KajKrone2003, Freund2020] or diploid reproduction [MohleSagitov2003, BirknerLiuSturm2018], using heuristics close to those of [MohleSagitov2001]. Finally some works use different techniques such as duality, e.g**C**: in the case of populations suffering recurrent bottlenecks [CasanovaMiroJegousse2020] or overlapping generations (Casanova et al

CAB

ACB

CBA

Selection 1

**A**: Noise introduces significant interference into the model, and we aim to ascertain whether attention consistency still holds under noisy conditions. By employing these strategies, we aim to ensure a comprehensive evaluation of our model across different learning scenarios. **B**: To comprehensively validate the effectiveness of attention-consistency learning, we introduced a noisy model, FedAvg-N, as another baseline**C**: Our experiments encompassed three distinct settings: (1) local learning: Each center independently trained its data; (2) global learning: aggregating datasets from all centers, along with training a multicenter centralized model; (3) federated learning: Training on data from all centers without sharing datasets, focusing solely on exchanging model parameters

BAC

CBA

ACB

CAB

Selection 2

**A**: In the experiments, we will demonstrate the importance of taking the alignment of data into consideration by comparing our method to these baselines.**B**: Solving DSBs is a subject of significant interest in recent years and has flourished in a number of different algorithms (De Bortoli et al., 2021; Chen et al., 2022a; Vargas et al., 2021; Bunne et al., 2023; Liu et al., 2022a)**C**: However, all these previous approaches focus on unaligned data, and therefore the methodologies all rely on IPF and are hence drastically different from ours

BCA

ACB

BCA

CAB

Selection 4

**A**: Here our aim is to investigate in the context of a toy model the consequences of density dependence that only affects development directly (fertility is affected indirectly, since it depends on the developmental stage of the individual)**B**: We do so for a one-dimensional i-state (i.e., the variable capturing the relevant differences among individuals ‘lives’ on the real line), so for an i-state space that comes equipped with an order relation**C**: In fact we shall assume that the presence of ‘larger’ individuals has a negative impact on the growth rate of ‘smaller’ individuals (as a motivating example one might think of trees and shading, with the i-state interpreted as ‘height’; but please note that we ignore spatial structure and that, consequently, the model is but a caricature). For the incorporation of space into physiologically structured population models see [27]. For an alternative approach to hierarchically structured models see [25].

CBA

ACB

BAC

ABC

Selection 4

**A**: Specifically, the procedure on the twin lattice was more than five times faster, requiring between 16%percent1616\%16 % and 20%percent2020\%20 % of the time required by the procedure on the model inclusion lattice. Furthermore, the proper implementation of the coherence principle allowed us to fit a much smaller number of models, ranging between 37%percent3737\%37 % and 54%percent5454\%54 % of the models fitted under the naive implementation of principle of coherence**B**: The point of main interest is the comparison in terms of efficiency, that we quantify with respect to the average execution time and the average number of fitted models. These can be found in the last two columns of Table 2 and, furthermore, the growth rates of these measurements are displayed in Figures 6 and 7. We can see that the procedure on the twin lattice is considerably more efficient**C**: It is also interesting to notice that the latter proportions appear to decrease as p𝑝pitalic_p increases. Table 2 also gives the average values over the 20 samples of the positive predicted value, the true-positive rate and the true-negative rate, both for the edges and for the colour classes of the selected graphs. These have satisfying values and there are not relevant differences between the two procedures, thereby showing that the increase in efficiency is not achieved at the cost of a lower level of performance of the selected model.

CBA

BAC

CBA

BCA

Selection 2

**A**: However, the current study is limited to single institutional data. In the future, we would like to explore the applicability of the method for multi-institutional data using domain adaptation techniques. **B**: This paper proposed a tensor learning-based pipeline for PAWP classification**C**: We demonstrated that: 1111) tensor-based features have a diagnostic value for PAWP, 2222) the integration of CM features improved the performance of unimodal and bi-modal methods, 3333) the pipeline can be used to screen a large population, as shown using decision curve analysis

BAC

CBA

CAB

Selection 4

**A**: Cat refers to concatenation, KP refers to Kronecker product. All omics and multimodal baselines were trained with the Reactome and Hallmark pathway sets. **B**: Best performance in bold, second best underlined**C**: Table 1: Results of SurvPath and baselines in predicting disease-specific patient survival measured with c-Index (at 20×\times×)

ACB

CBA

BCA

ABC

Selection 2

**A**: Most notable is the effect of such systems to smooth and provide an additional degree of control over external noise, consequentially increasing resilience and robustness. The inclusion of feedback and feedforward loops can enhance this effect, providing systems with additional degrees of control and contributing to so-called perfect adaptation [46, 45]**B**: Such results potentially explain the complexity in the replication network structure seen in some viruses; for example, in the human adenovirus [37]. Indirectly, these loops (and by extension, more complicated network structures) provide systems with the ability to tune the first passage time distribution, potentially yielding an optimal lysis time [47, 13]. While we restrict our analysis to a single non-local feedback or feedforward, future work may study more general optimal network structures informed by more specific biological problems.**C**: Such loops provide a potential explanation for out-of-order progression in some systems, for example, whereby viral replication does not occur as a unidirectional process [2]. Our work demonstrates that feedback loops could yield a fitness advantage through more favourable statistical properties of viral load compared with perfect progression through the replication cycle

CAB

CBA

ACB

BCA

Selection 3

**A**: Elman networks, more commonly known as vanilla recurrent neural networks (RNN), attempt to introduce the concept of a time-dependent dynamic memory [16]. The idea is to make predictions about inputs based on contextual information**B**: Context-based predictions can be done for four input-output schemes: one-to-one, one-to-many, many-to-one, and many-to-many. One-to-one models are a variation of a classic neural network, one-to-many models are best for image caption generation, many-to-one models are best for sentiment analysis, and many-to-many models are best for translation or video frame captioning**C**: Fig. 1 is an example of the basic structure of a vanilla RNN.

CBA

BCA

ABC

BAC

Selection 3

**A**: 1). The three regimes correspond to the network topology updates being much faster (annealed regime), comparable (intermediate regime) or much slower (quenched regime) than the spread of the disease. **B**: The analysis of epidemics on time-variant networks is often under the assumption of timescale separation: either the network changes significantly faster than the spread of the disease or the epidemic evolves significantly faster than the topology updates of the network**C**: Timescale separation of the network and the epidemic results in three regimes (Fig

CAB

BCA

ACB

Selection 1

**A**: We explore features of covers arising from networks, and characterise many of the familiar classes in terms of properties of their associated covers. It is to be hoped that encoding network properties in the properties of sets of sets will enable some new directions to be pursued in studying phylogenetic networks. **B**: They have been shown to correspond to a set of covers of finite sets that satisfy a property called “expanding”**C**: The class of labellable networks contains many commonly studied classes

ABC

CBA

CAB

Selection 3

**A**: Suppose we are given a sample from the selected locus at the present time t=0𝑡0t=0italic_t = 0, and that we know the allelic types of the sample but we do not know how the sample was produced. What is the genealogy of the sample? This question was answered by Barton et al**B**: For samples from a population at stationarity as in Section 1, Barton et al. (2004) proved that this could be done rigorously starting with a Moran model with finite N𝑁Nitalic_N then passing to the diffusion limit. Barton and Etheridge (2004) explored some properties of gene genealogies under this model, and Etheridge et al. (2006) used the same idea to describe genetic ancestries following a selective sweep. **C**: (2004), who modeled the ancestral process using the structured coalescent with allelic types as subpopulations. The structured coalescent can be a model of subdivision with migration between local populations (Takahata, 1988; Notohara, 1990; Herbots, 1997) or a model of selection with mutation between allelic types (Kaplan et al., 1988; Darden et al., 1989)

CBA

ACB

BCA

ABC

Selection 2

**A**: MIXALIME also provides machinery to test for the differential allele-specificity between two sample groups (i.e**B**: control and test)**C**: We employ Wald or likelihood-ratio test (LRT) to see if there is a difference in parameters estimates between two groups:

BCA

BAC

ABC

BAC

Selection 3

**A**: fully susceptible (σ⁢(a)=1𝜎𝑎1\sigma(a)=1italic_σ ( italic_a ) = 1)**B**: If TR=∞subscript𝑇𝑅T_{R}=\inftyitalic_T start_POSTSUBSCRIPT italic_R end_POSTSUBSCRIPT = ∞ a.s., individuals are permanently immune following an infection**C**: Depending on the distribution of TRsubscript𝑇𝑅T_{R}italic_T start_POSTSUBSCRIPT italic_R end_POSTSUBSCRIPT, we can recover either the SIR, SIS or SIRS model

CBA

BCA

ACB

BCA

Selection 3

**A**: In particular, the analytical species abundance distribution derived from the DMFT follows the Gamma distribution, a widely utilized probability distribution in macroecology [32, 1]**B**: In other words, due to the non-linear nature of the corresponding Fokker-Planck equation and known pathologies in the GLV model (also observed in the quenched case), the dynamics may not converge to the stationary solution, leading to divergent trajectories.**C**: Again, similar truncated fat-tailed distribution has been recently shown in the chaotic phase [36] and in the strongly interacting limit [5, 41] of the QGLV with immigration. We have successfully obtained the phase diagram for the case of annealed white noise (AWN), and numerical simulations for the case J⁢(x)=x𝐽𝑥𝑥J(x)=xitalic_J ( italic_x ) = italic_x have revealed the potential for unbounded growth when the initial conditions possess large values, despite the existence of an analytically stationary solution

BAC

ACB

CAB

Selection 3

**A**: Transcription—the synthesis of RNA—is typically modelled as a multistep process in which the gene switches between multiple states before it eventually produces an RNA molecule**B**: Depending on the level of details, transitions between gene states may reflect individual biochemical events, such as binding of transcription factors and RNA polymerase at the promoter, or more phenomenologically, a combination of these events that results in the gene being either active or inactive**C**: Once the RNA is produced, it goes through a series of steps until it is eventually degraded. In the Markovian setting, these steps can be described by the following reaction scheme

CBA

ABC

BAC

CAB

Selection 2

**A**: Ho et al**B**: (2018) calculate transition probabilities for the birth/birth-death process—a restricted bivariate case where the death rate of one type vanishes, but rates may be otherwise nonlinear. Xu et al**C**: (2015) use branching process approximations of birth-death processes and generating-function machinery (Wilf, 2005) for moment estimation.

CAB

BAC

ABC

BCA

Selection 3

**A**: Where the similarity represents the highest alignment score between the amino acid sequences of the test and train sets using BLAST identity**B**: Furthermore, in Figure 2 we report the performance of all Prot2text models with respect to different similarity thresholds**C**: We observe that for test proteins with low similarity scores with the train set (between 20% and 30%) and for proteins with no counterpart in the train set, the Prot2TextMEDIUM is the dominant one while for higher similarity scores Prot2TextLARGE performs better.

BCA

BAC

BCA

CBA

Selection 2

**A**: strength [cf**B**: Fig. 3(b) and discussion after Eq. (31)]**C**: Note that the saddle point 𝑸∗⁢(𝑫=𝑫¯)superscript𝑸𝑫¯𝑫\boldsymbol{Q}^{*}\left(\boldsymbol{D}=\overline{\boldsymbol{D}}\right)bold_italic_Q start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT ( bold_italic_D = over¯ start_ARG bold_italic_D end_ARG )

BCA

CBA

ABC

CBA

Selection 3

**A**: As shown in Figure 7, changes in the shape of the reconstructed volumes indicate a large flexible motion in the continuous conformational space. The corresponding predicted volumes and the distribution of predicted latent z𝑧zitalic_z along the first two principal component axes are similar to that from the literature [30].**B**: For experimental datasets, the distribution of noise is often more complex than simulated datasets**C**: Based on our simplified noise estimation method (Equation 9), the SNR of spliceosome datasets is around −11.711.7-11.7- 11.7 dB, which is worse than the simulated datasets above. We predicted the latent z𝑧zitalic_z for the full dataset, then sampled three points along the first component axis and generated the corresponding volumes

ABC

BAC

CAB

ABC

Selection 3

**A**: ML, VG, MS, SJR were supported by SNSF grants CRSK-3_190526 and 310030_204938 awarded to SJR. E**B**: Upschulte and T**C**: Dickscheid received funding from Priority Program 2041 (SPP 2041) “Computational Connectomics” of the German Research Foundation (DFG), and the Helmholtz Association’s Initiative and Networking Fund through the Helmholtz International BigBrain Analytics and Learning Laboratory (HIBALL) under the Helmholtz International Lab grant agreement InterLabs-0015. The authors gratefully acknowledge the computing time granted through JARA on the supercomputer JURECA at Forschungszentrum Jülich.

ACB

BAC

ABC

CBA

Selection 3

**A**: A careful examination of these explicit terms may be of interest in future work, but we collapse them into single functions and continue these calculations with the aid of a symbolic package up to some chosen order k𝑘kitalic_k: **B**: Three-body interactions are apparent, as are the existence of interactions between coupling and heterogeneity**C**: These terms, with b=0𝑏0b=0italic_b = 0, are the second-order interaction terms from [69]

ACB

BAC

CAB

ACB

Selection 3

**A**: In Sec. III, for the case of noninteger α𝛼\alphaitalic_α, we construct a first-order perturbation theory for the effect of noise on the characteristic function and derive macroscopic observables: population-mean voltage and firing rate**B**: In Sec. IV, the theoretical results for macroscopic states of homogeneous populations of QIFs are reported. In Sec. V, we compare the theoretical results against the background of the results of numerical simulation (also for a less mathematically challenging case of heterogeneous populations) and discuss general implications of the theoretical results and the examination (Appendix C) of the possibility to construct the generalization of the pseudocumulant expansion to the case of noninteger α𝛼\alphaitalic_α**C**: In Sec. VI, conclusions are drawn.

CBA

ABC

BCA

BAC

Selection 2

**A**: Predictive coding can be performed over any sensory modality that has some temporal sequence**B**: As natural speech forms a cognitive map, predictive coding may underlie the geometry of human language**C**: Intriguingly, large language models train on causal word prediction, a form of predictive coding, build internal maps that support generalized reasoning, answer questions, and mimic other forms of higher order

BAC

ABC

ACB

CBA

Selection 2

**A**: Temporal-spatial convolution is used with spatial modules, made with self and graph attention, to reveal spatial features of brain activity**B**: The linear layer is used to project the feature dimension.**C**: Then, the model obtains results by matching test data to templates. (B) Architecture of the EEG encoder

CBA

ACB

BCA

BAC

Selection 3

**A**: Even with these simplifying assumptions, species tree estimation is confounded by gene tree heterogeneity. Heterogeneity is particular problematic for concatenation-based methods, as the species tree for the entire concatenated sequence can disagree with gene trees for particular loci, Roch and Steel (2015)**B**: Many theoretical results, positive and negative, have been established when the only source of heterogeneity is ILS, see Degnan and Rosenberg (2006), Allman et al. (2011), and Mirarab et al. (2014). Incomplete lineage sorting is modeled by the multispecies coalescent (MSC) model.**C**: Common sources of heterogeneity include incomplete lineage sorting (ILS) Rannala and Yang (2003), horizontal gene transfer (HGT) Roch and Snir (2013), and gene duplication and loss (GDL) Arvestad et al. (2009)

BCA

ACB

ABC

CBA

Selection 2

**A**: In fact, Lawley and Madrid [24] have extended these works to also consider the effect of small noise**B**: Relevant to our paper, they argue that the relative benefit of increased numbers of walkers is relatively weak (scaling as 1/log⁡(N)1𝑁1/\log(N)1 / roman_log ( italic_N )), and so it remains unclear if many biological systems are in this regime**C**: We also note the work of Weiss [42], who determined first-hitting-time estimates when the walkers are initially uniformly distributed throughout the domain.

BAC

ACB

ABC

Selection 4

**A**: A recent work demonstrated that the weight uncertainty with the form of SaS structure can be also incorporated into the transformer [45]**B**: In addition, gated recurrent neural networks with multiplicative mechanisms were recently shown to be able to learn to implement linear self-attention [46]**C**: Furthermore, the relationship between linear transformers allowing for faster autoregressive learning and RNNs was established in a recent work [47]. Taken together, our current work would be a starting point to establish the bridge between the biological learning (towards the science of specialized brain circuits) and transformer learning within the seminal predictive coding hypothesis, which can be put in the theoretically solid variational free energy minimization conceptual framework.

CAB

BCA

BAC

ABC

Selection 4

**A**: Data acquisition**B**: The images were captured using Leica DMi8 microscope (Leica) equipped with 10×/0.32 objective lens. We obtained one whole slide image from each group.**C**: Bright-field images used in this paper were obtained under the protocol described in [5]

ABC

BAC

CAB

ACB

Selection 4

**A**: In simple terms, an ANN learns from data by adjusting the weights of the connections between nodes in order to minimize a loss function that measures the difference between the desired output and the actual output of the network. More specifically, the optimization step is performed using the Stochastic Gradient Descent (SGD) algorithm, which iteratively updates the weights of the network by moving them in the direction of the steepest descent of the empirical loss function of a single training sample**B**: Such a mechanism does not seem to be biologically plausible for BNNs, as many authors have pointed out. Parameter update in BNNs occurs only locally, and distant neurons are only indirectly connected through the endogenous reward system. This observation is closely related to the weight transportation problem [6, 2, 4]. We refer to [12, 11] for a detailed discussion about the role of SGD in BNN, which the author of [10, Section 5] summarizes as follows:**C**: The gradient itself is computed with the so-called backpropagation algorithm. In particular, the update of any parameter is based on the states of all other parameters

ACB

BAC

ABC

BAC

Selection 1

**A**: The formation of NCNM was formed by a self-assembly of homogeneous nanoparticles with negative surface charge in the desired shallow channel using Laplace pressure to halt the solvent at the base and evaporation of solvent. A close-packed fcc was formed by the growth of the ordered lattice induced by the evaporation.**B**: The fabrication of microchannel and formation of the NCNM for the fluidic memristor is similar to previously reported methods [36, 37] and is described in the SI in detail. A master for multi-layered channels (target heights are 5 μ𝜇\muitalic_μm for shallow channel and 100 μ𝜇\muitalic_μm for deep) was created using a multi-step UV exposure with negative photoresist (PR, SU-8 2005, 3050, Microchem Co., USA)**C**: After surface treatment of the master with (3,3,3-trifluoropropyl)silane (452807, Sigma-Aldrich, USA) for easy separation, Polydimethylsiloxane (PDMS, Sylgard, Dow Corning Korea Ltd., Korea) was poured and cured by heating. The detached PDMS device was bonded with a slide glass

CBA

CAB

ABC

ACB

Selection 2

**A**: It offers an extensive description of the trade-off between disease control and intervention costs applied in an infectious disease model and can be additive. Based on the considerations mentioned earlier, the following OCP is formulated by a hybrid cost function combining linear and quadratic terms, where the controls are shown in red for emphasis.**B**: Different types of cost functions may be appropriate for different scenarios, such as an exponential cost function avusuglo2023leveraging ; grandits2019optimal , where it can be used if the decision-maker under uncertainty, and its concave nature of exponential cost describes a risk-averse attitude, or a quadratic cost function, which its convexity ensures that any local minimum is also a global minimum**C**: Besides, the choice of the cost function is contingent upon the specific goals of the optimization problem

BCA

CAB

ACB

CBA

Selection 4

**A**: They control voluntary movement and also make a crucial role in cognitive processes (e.g., action selection)**B**: Dysfunction in the BG is related to movement disorder (e.g., PD) and cognitive disorder.**C**: The BG exhibit diverse functions for motor and cognition

BCA

ABC

ACB

CBA

Selection 1

**A**: Therefore, future work should aim to compute and store the Rashomon set of a wider variety of model classes. Future work may investigate incorporating Rashomon sets that may be well-approximated (e.g., GAMs, [10]), but not computed exactly, into the RID approach. Nonetheless, sparse trees are highly flexible, and using them with RID improves the trustworthiness and transparency of variable importance measures, enabling researchers to uncover important, reproducible relationships about complex processes without being misled by the Rashomon effect. **B**: RID can be directly computed for any model class for which the Rashomon set can be found – at the time of publishing, decision trees, linear models, and GLMs**C**: A limitation is that currently, there are relatively few model classes for which the Rashomon set can be computed

CAB

BAC

ABC

Selection 1

**A**: We report our performance against five leading methods [24, 41, 15, 32, 37]**B**: We detail our experimental methodology in Sec. 5.1 and report quantitative and qualitative evaluations in Sec. 5.2. Ablation studies are presented in Sec. 6. **C**: Following common practice in the field, we evaluate DREAM with the largest public neuroimaging dataset, the Natural Scene Dataset [2]

CBA

BCA

BAC

ABC

Selection 2

**A**: We note that the optogenetic treatment could have benefits in comparison to the traditional deep-brain-stimulation (DBS) treatment**B**: The DBS has the following disadvantages OG3 ; OG4 ; (a) it is difficult to accurately determine the target cells, leading to cause many side effects and (b) a process with many trial and errors is necessary to each patient for optimal control**C**: On the other hand, the target cells can be accurately located by optogenetic stimulation. Hence, side effects and trial-and-error process may be decreased in spite of limitations for application to the human patients OG3 .

CAB

CBA

ABC

Selection 4

**A**: Bürger and Lynch, 1995; Gomulkiewicz and Holt, 1995; Matuszewski et al., 2015). For models like this, our findings may be of relevance either for the initial response (in case of a sudden shift) or even in the long term (in case of a moving optimum). Below, we briefly review recent treatments focusing on the polygenic dynamics underlying trait evolution. **B**: Because we assume exponential directional selection, beneficial mutations, and weak random genetic drift, all mutants that become ‘established’ will sweep to fixation in the long run. Therefore, we focus on the initial response when characterizing patterns of adaptation as sweep-like vs. shift-like**C**: In Section 6.4, we argue that our results are of relevance for other modes of directional selection. A popular model is one in which a population is initially in equilibrium under mutation-stabilizing selection-drift balance and then starts to evolve due to a sudden or a continuous change in the environment, modeled by a sudden shift in the optimum phenotype or by a continuously moving optimum (e.g

CAB

ACB

CBA

BAC

Selection 1

**A**: In this study, we aim to enhance GNNs to better capture long-range interactions**B**: This novel approach reduces interaction distances between nodes to a single hop. Extensive experiments on four long-range graph benchmarks validate our method’s ability to enhance any GNN to capture long-range interactions. **C**: We achieve this by transforming original atoms into neural atoms, facilitating information exchange, and then projecting the improved information back to atomic representations

BAC

BCA

ACB

BAC

Selection 3

**A**: In one set of experiments, in order to study dependence on model size, we take a selection of LLMs from the Pythia family [22]**B**: We provide full details of the experiments in the Supplementary Information. Throughout the paper we primarily investigate the open source GPT-J model with around 6 billion parameters [5]**C**: The code for all experiments is provided at github.com/rmldj/memory-llm-paper.

ACB

BAC

ABC

Selection 3

**A**: Our analysis has focused only on reconstructing the covariance between the taxa in a set of samples**B**: To understand more general relationships, including across kingdoms, other methods should be employed alongside covariance network reconstruction**C**: In particular, non-negative tensor factorization methods show promise in identifying relationships between groups of taxa. Two popular tools, MOFA+[37] and DIABLO[38] are both designed to use non-negative tensor factorization to identify groups of related variables across data types.

CAB

BCA

ABC

BCA

Selection 3

**A**: The virus is now managed similarly to seasonal influenza, with new vaccine booster shots tailored to target emerging virus variants. **B**: At present, daily life in most countries has returned to pre-pandemic norms**C**: The development and widespread use of vaccines and treatments eventually brought an end to the pandemic, officially declared by the World Health Organization (WHO) on May 5, 2023[1]

CBA

BCA

ACB

ABC

Selection 1

**A**: Since each p𝑝pitalic_p-dimensional simplex corresponds to a collection of (p+1)𝑝1(p+1)( italic_p + 1 ) data points222Recall that points are 00-dimensional simplices, edges are 1111-dimensional, etc., we can assign weights to the datapoints themselves by considering for each node in the Vietoris-Rips complex, the sum of the weights of its cofaces**B**: For example, given a 1111-dimensional harmonic representative, we can translate the weights from the edges to the nodes by assigning to each node the sum of the weights of its adjacent edges**C**: This give us a set of weights on the nodes for each bar.

CAB

BCA

CAB

ABC

Selection 4

**A**: In iopart.cls, to make any line start at the left margin of the page, add \fl at start of the line (to indicate full left).**B**: The iopart.cls class file automatically does this and indents each line of a display**C**: The normal style for aligning displayed equations in our published journal articles is to align them left rather than centre

ACB

BAC

CBA

ACB

Selection 3

**A**: As evidenced in Fig**B**: However, the pre-trained LMs are not inherently trained on data specific to the target environmental ecosystems, which often results in a failure to effectively capture feature dependencies within the descriptions**C**: 3 (b), the input features from samples in different seasons tend to be mixed in the latent embedding space of the original pre-trained DistilBERT model. This indicates that a standard LM, when applied directly, may fall short of capturing semantics from the text generated for our target task.

CAB

BAC

ACB

Selection 3

**A**: Nucleotide Transformer introduced the first large-scale transformer-based DNA LMs. All NT models share the same architecture, but differ in their number of training genomes and model parameters**B**: A second generation of multispecies models released later (NT-V2) extended the input length to 12,282 bp. The NT models were evaluated on tasks comprising promoter, SS, histone modification and enhancer prediction with a context length of up to 600 bp. **C**: Models were trained on either the human reference genome, 3,202 different genetically diverse human genomes or a selection of 850 genomes from a range of species. To increase the receptive field of the model, sequences were tokenized as 6-mers, allowing for processing sequences of up to 5,994 bp in length

ABC

ACB

BAC

Selection 3

**A**: However, from a statistical perspective, concerns arise regarding the potential for false discoveries of genes that lack genuine spatial variability. This concern becomes more pronounced when a large number of genes are simultaneously tested across most frameworks. If the false discovery rate or type 1 error is not adequately controlled, it may lead to incorrect conclusions and the selection of numerous genes that exhibit false spatial variability.**B**: The mathematical models employed for capturing the data generation process and the innovative model-free SVG detection technique have proven valuable for uncovering significant SVGs that offer critical biological insights**C**: We have previously discussed both model-based and model-free methods for detecting SVGs

BAC

BCA

CBA

Selection 4

**A**: As shown in Figure 2, MolIG, which utilizes both pre-training and data augmentation strategies (represented by the grey bar), performs best among all model architectures**B**: Models without pre-training (represented by ’w/o pretrain’) perform worst in all tasks. Pre-training strategies without data augmentation (represented by ’w/o img aug’) perform second best, yet they show a significant performance increase compared to models without any pre-training. Excluding either of these two components can easily lead to a decrease in performance.**C**: This ablation study aims to evaluate whether pre-training strategies and data augmentation strategies can help the model achieve better performance in downstream tasks

ACB

BCA

CAB

ABC

Selection 2

**A**: During the downstream task, the classification head is attached to the pretrained encoder and its weights are fine-tuned on the dataset of said task in a fully-supervised manner. The architecture of the ”downstream model” is used as a baseline in all of our experiments, where all of its layer weights are adjusted during fully-supervised training. **B**: The weights of the encoder or ”backbone network” are frozen after the pretraining phase**C**: Figure 2: The architecture of the CNN encoder trained via SSL contrastive learning (left) and the classification head trained for the downstream task (right)

CBA

ACB

BAC

ABC

Selection 1

**A**: Our main analysis encompasses two distinct definitions of tumor recurrence: a broad definition including edema, enhancing core, and necrotic core, and a more specific definition aligned with current RANO guidelines focusing only on the enhancing core**B**: Furthermore, we conduct a patient-specific comparison with the Standard Plan, classifying a model as ’Better’ if it provides higher coverage for an individual patient and ’Worse’ if it falls short of the Standard Plan’s coverage. In instances of equal coverage (e.g., both achieving 100%), we label the outcome as ’Equal’. The comparison outcomes, including average results, are depicted in Figures 5c and 6c. **C**: Figures 5 and 6 present the mean and standard deviation of Recurrence Coverage for these definitions

BCA

CAB

ACB

CAB

Selection 3

**A**: Figure 2: Time course of C𝐶Citalic_C derived by solving Eqs. (3)–(4) for the same parameter values as in Fig. 1**B**: In (a) and (b) we show the time evolution for two initial conditions to illustrate that the only fixed point of the system is stable but excitable**C**: In (c) there is a stable limit cycle.

ABC

BAC

BCA

BAC

Selection 1

**A**: The top three values for each metric and each track are marked as gold, silver and bronze cells in decreasing order. A ‘*’ behind the team name means that the segmentation predictions have been converted from the multiclass submissions and inserted here. If a team only submitted to one of the tracks the columns of the other track are filled with a ‘-’.**B**: Table 4: Binary segmentation task results (mean ±plus-or-minus\pm± standard deviation) for MRA and CTA in Dice scores, centerline Dice (clDice) scores and errors in the zero-th Betti number β0subscript𝛽0\beta_{0}italic_β start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT**C**: The arrow indicates the favorable direction

CAB

BAC

ACB

Selection 1

**A**: Building upon DDPMs, MoFusion (Dabral et al. 2023) employs U-Net (Ronneberger, Fischer, and Brox 2015) as the backbone for the denosing kernel in motion sequence synthesis. Apart from applications on continuous data, many research efforts are devoted to discete data generation**B**: EDP-GNN (Niu et al. 2020) introduce Score SDEs, a learning paradigm of diffusion models, to the generation of discrete graph adjancency matrices.**C**: The formation of human motion skeletons share similarities to molecules, as both are represented by point clouds connected by edges. The difference is that human motion generation does not require predictions of edges

ABC

ACB

BAC

BCA

Selection 4

**A**: In this diagram, every point is represented by a cell enclosed by the red solid lines and dashed curves, then the surface area can be calculated with such information.**B**: The purpose of this step is to obtain the dual complex 𝒞𝒞\mathcal{C}caligraphic_C and the conjugated Laguerre diagram**C**: As shown in Figure 1, the black points are the center of atoms and dashed lines connecting black points denote the edges in the tetrahedrization

BAC

BCA

CBA

BAC

Selection 3

**A**: 2021) and ChemBERTa-2 (Ahmad et al. 2022). **B**: Additionally, in our experiments, we tested two types of M-Encoder: CHEM-BERT (Kim et al**C**: Since any molecular pretraining model can serve as the M-Encoder in our method, in this section we mainly describe the Transformer encoder block (TEB), the fundamental component composing mainstream language models, as well as the internal structure of the MT-Encoder

ACB

CBA

BCA

ACB

Selection 2

**A**: Of course, this qualitative understanding is no substitute for the detailed analysis obtained by solving the replicator equation and, in particular, cannot predict the existence of an all-cooperators regime. **B**: Direct application of the variance decomposition formula Weiss_2005 shows that a non-degenerate distribution of group sizes always increases the variance of a cooperator’s fitness**C**: A fundamental result of Wilson’s approach is that cooperation is promoted by increasing the variability of group composition or, if fitness is not a linear function of the number of cooperators, by increasing the variance of a cooperator’s fitness

CAB

CBA

ACB

Selection 3

**A**: In exploring the influence of racial and ethnic groups on BP values, we categorized the final pre-processed dataset based on the racial/ethnic groups listed in Table II**B**: Importantly, race and ethnicity are compound factors, impacted by biology (genetic factors that impact BP from a physiological or anatomical standpoint) and also by social determinants of health (factors related to race/ethnicity disparities)**C**: Table II presents the mean and STD for DBP and SBP across racial and ethnic groups. Accordingly, the African American or Black group had the highest mean SBP and DBP values. The Asian population demonstrates the lowest SBP, while the Hispanic group has the minimum DBP.

BAC

BCA

ACB

ABC

Selection 4

**A**: The effects of repressing the clock can be seen in Figure (4A) which shows the dynamics of the core circadian oscillator gene BMAL1. The failure of BMAL1 to oscillate indicates the repression of the circadian clock. Figure (4B) shows the filter output for BMAL1. **B**: In order to demonstrate the scalability of our approach on real world datasets, we used the PKF to analyze time-course gene expression measurements (RNAseq) of a circadian clock dataset in human cancer cell lines [rebekah_clock] with contains of roughly 1.8 million gene expression measurements consisting of 32337 gene expression measurements across two conditions and 14 time-points with two replicates per time-point and condition. Briefly, the mammalian circadian clock is an cell-autonomous oscillatory genetic circuit which regulates diverse cellular functions by synchronizing them to daylight patterns [takahashi], [sgolden]**C**: For a brief description of the mammalian clock, refer to Appendix A.5.2. In this dataset, the circadian clock is disrupted by activating the gene MyC [rebekah_clock] which causes the clock to be repressed

CAB

CBA

Selection 1

**A**: We propose Absorb-Escape, a generalisable post-training algorithm for refining the quality of generated discrete sequences**B**: We show that Absorb-Escape further increases the performance of DiscDiff by 4% in long DNA generation**C**: In addition, Absorb-Escape allows control over the property of generated samples.

ACB

CAB

ABC

BAC

Selection 3

**A**: FM-indexes over minimizer digests are known to be usually significantly faster than indexes over the original datasets, both because some characters are not represented in the digests and because we use a backward step for each minimizer rather than for each character, incurring fewer cache misses**B**: If we are willing to sacrifice the true-positive rate moderately, we can increase w𝑤witalic_w and also achieve significant speedups**C**: Interestingly, however, we achieved slightly better speedups with both kernelization and minimizers than with either separately.

ABC

BAC

CBA

ABC

Selection 2

**A**: We base our kernels on the vLLM implementation ([51]).**B**: Fused kernels enable complex operations, such as Rotary Embedding ([50]) which would otherwise be executed sequentially, to be combined into a single kernel than can be executed in parallel on a GPU**C**: To optimize the forward pass, we implement fused CUDA kernels

CBA

BCA

ACB

Selection 1

**A**: This feature denotes the level of light scatter at a 90-degree angle in relation to the laser beam**B**: Side Scatter (SSC)-Cell’s granularity**C**: SSC reflects the internal complexity or granularity of the cell, encompassing features like the presence of granules, nuclei, or other organelles. A high importance of this feature may suggest that either our model can differentiate between cells that have different degrees of complexity, such as lymphocytes, monocytes, and granulocytes, or our model can identify cells that have abnormal granularity, such as blast cells or malignant cells.

BAC

ABC

ACB

CAB

Selection 1

**A**: The natural death**B**: both phases of the pandemic**C**: The recovery periods σ−1superscript𝜎1\sigma^{-1}italic_σ start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT and γ−1superscript𝛾1\gamma^{-1}italic_γ start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT are taken to be 21⁢days21days21\,{\rm days}21 roman_days

CBA

BAC

CAB

BCA

Selection 3

**A**: Parameter reallocation within the network also plays an important role in memory conservation**B**: This step involves redistributing parameters by expanding channel width in critical modules while reducing it in less important ones**C**: This strategy allows for more efficient learning of representations in higher-dimensional spaces without losing feature information, while also speeding up inference by removing redundant parameters.

CAB

BCA

ABC

CAB

Selection 3

**A**: Any approximation to 𝐊𝐊\mathbf{K}bold_K that accounts for identical repetitions, including all known lossless statistical compression algorithms, can achieve equivalent or superior results, as demonstrated in the Supplementary Information and [4]**B**: This alignment with ‘Template Program A’ effectively highlights the association of AT with well-established principles of compression and coding theory, thereby refuting the initial claim of its authors to present a unique methodology**C**: Moreover, the authors’ suggestion that their index may be generalised as a universally applicable algorithm for any object (including text) [31] further underscores the

CAB

ABC

CBA

BAC

Selection 2

**A**: This tactic for Cognitive Security against readout requires alteration, albeit strategically minimal alteration. And as seen with the similarities between SMON/SION and MOA/IOA, there is a meaningful analogy between AA and defensive measures against RA, and between RA and defensive measures against AA. If no readout is made of cognition and neural activity, then alteration may not in general be reliably detected**B**: Finally, there is an interesting notion which emerges from the discussion of attack and defense methods. The Bonaci et al BCI Anonymizer is an innovative approach to the proper design of BCI hardware, but remains vulnerable to air-gap attacks, malicious implants, or BCI hardware vulnerabilities. In order to maximally mitigate such vulnerabilities, one would likely need to induce physiologic noise**C**: The roles of the attacker and the defender exist as odd sorts of mirror images in both cases. While we do not currently see any path towards verifying this one way or another, it leads us to make the following conjecture. There may be no means to achieve physiology-level security guarantees against readout without enacting some degree of alteration to cognition itself, nor such guarantees against alteration without performing some readout. As with much of the myriad implications of this paper and the problems it seeks to begin addressing, there is a great deal of future work to be done on the topic.

CAB

BAC

BCA

Selection 2

**A**: These metrics are evaluated both within a batch of 64 samples and across the entire test set, following the methodology outlined in Liu et al. (2023b). As presented in Table 1, MolBind’s performance surpasses that of MolCA by a large margin. Overall, MolBind significantly outperforms prior work on two molecule-language retrieval tasks, validating its effectiveness in aligning the molecule and language modalities. **B**: We evaluate the graph-language and conformation-language retrieval performance on the MoBind-M4 dataset. For our baselines, we employ Sci-BERT Beltagy et al. (2019), KV-PLM Zeng et al**C**: (2022), MoMu Su et al. (2022), and MolCA Liu et al. (2023b), assessing their performance using Recall@1 and Recall@20

CAB

ACB

BAC

Selection 1

**A**: If more than one route is found for a molecule, the route scored the highest by AiZynthFinder is retained. For solved routes, we compute the depth of the tree. Since the maximum tree depth allowed by AiZynthFinder’s default parameters is 7, we assign a depth of 10 for molecules where a route is not found. **B**: We use AiZynthFinder [11] with default parameters to compute the routes of both datasets**C**: A route is considered to be solved if all leaves are purchasable molecules

ACB

CAB

ABC

Selection 2

**A**: The three panels summarize the system behavior for α=0.8,0.4𝛼0.80.4\alpha=0.8,0.4italic_α = 0.8 , 0.4, and 0.2, where the inner invasion in odd-labeled loop is strong, intermediate or weak**B**: Our first results, shown in Fig. 3, were obtained when the interaction between the triplets are strong**C**: Evidently, the probability α+δ𝛼𝛿\alpha+\deltaitalic_α + italic_δ cannot exceed 1, which explains the different maximal δ𝛿\deltaitalic_δ values for each panels.

BCA

BAC

BCA

ACB

Selection 2

**A**: We compare several transformer architectures on the task of cancer detection on a single ROI**B**: We choose a modified ResNet18 [9] as our ROI-scale baseline, with only one sequence of convolutions and batch normalization in each residual block. This reduction in the number of parameters mitigates overfitting and is associated with an increase in performance[2]. **C**: To that end, we use a standard Vision Transformer[6] (ViT) architecture, a Compact Convolutional Transformer[7] (CCT), and a Pyramid Vision Transformer[8] (PvT)

BCA

CAB

ACB

CBA

Selection 3

**A**: One response was removed because the person stated in the open ended comments that they did not use the tool, but filled out the survey anyway. User responses indicated that the dashboard were visually appealing (>89% appealing or very appealing) sufficiently easy to navigate (>82% responded easy or very easy), provided a positive user experience (>91% satisfied or very satisfied), and trusted the accuracy of the data (>86% trust or strongly trust). **B**: Surveys were sent to 78 users of the Grafana v3 dashboard to evaluate experiences with a 35.89% response rate (n=28)**C**: The GEMS Sensing user interface evolved from simple HTML and CSS pages (Figure 4A) into a REACT.JS frontend (Figure 4D) and then into a Grafana-based set of organizations and dashboards (Figure 4G)

BCA

ABC

CBA

BAC

Selection 3

**A**: These MLMs are available in the HuggingFace transformers library Wolf et al. (2020). However, it can be noted that some level of adaptation was needed at the tokenization step to use some of the models within NLstruct, especially for frALBERT and FlauBERT models555The source code is available here: https://gitlab.lisn.upsaclay.fr/nlp/deep-learning/nlstruct. **B**: To compare French MLMs, we train one NER model using each of the language models in our benchmark**C**: The text encoder component in NLstruct relies on embeddings produced by a BERT language model, a char-CNN encoder, and static French FastText embeddings

BAC

ABC

CBA

CAB

Selection 3

**A**: The proposed method has a wide variety of potential applications, including drug discovery, and molecular design**B**: By creating low-dimensional embeddings and using them to find molecules with related qualities, the approach makes it possible to construct unique compounds with certain properties. Overall, this signifies a promising avenue for molecular structure analysis, employing kernel methods to unlock new possibilities. Following are our contributions:**C**: It offers the opportunity to quickly search through vast datasets of chemical structures in search of compounds with desirable properties

BCA

ABC

CBA

ACB

Selection 4

**A**: Specifically, we first introduce indices evaluating signal distortion by the gain and the phase delay characteristics and derive these characteristics of MC channels based on the diffusion equation and the rate equation**B**: We then show design conditions for MC channels in which the magnitude of distortion becomes below a specified level based on the indices. Using the proposed method, we demonstrate the design procedure of specific MC channels that satisfy given specifications. Finally, the roles of MC channels in nature are discussed from the perspective of signal distortion.**C**: In this paper, we propose a method to analyze signal distortion caused by one-dimensional diffusion-based MC channels

CBA

BAC

BCA

CAB

Selection 3

**A**: The prediction-based method, a learning-based approach, employs a deep neural network to extract meaningful features from past states**B**: It leverages the neural network’s ability to automatically learn and remember states without requiring manually designed storage solutions and also promotes more efficient exploration**C**: However, this approach cannot effectively encourage exploration as RL training progresses, particularly when the agent needs to escape out of local optima. Specifically, when the neural network exhibits unexpectedly high generalization performance, it may predict low errors even for unvisited states and hinder the agent from exploring the space further.

ACB

CAB

ABC

BAC

Selection 3

**A**: 2023b; Zhang et al. 2022b)**B**: Recently, a growing body of literature has explored the integration of SNNs with DRL (Zhang et al. 2022a; Wang et al**C**: Several methods convert a trained DQN into a SNN version (Patel et al. 2019; Tan, Patel, and Kozma 2021) or directly train a deep spiking Q-network (Liu et al. 2021; Chen et al. 2022). These endeavors yield competitive performance on Atari video games with discrete action spaces compared to the original DQN.

BAC

BCA

CAB

ACB

Selection 1

**A**: We remove the data augmentation during the second training stage, which means training solely using the DFS order that can generate canonical SMILES**B**: Data Augmentation**C**: Table 6 demonstrates a significant decline in model performance across all metrics. This clearly demonstrates that our data augmentation significantly improves the model’s performance.

BCA

BAC

ACB

Selection 2

**A**: For flow-based models, despite exhibiting high uniqueness and novelty, their validity was lower, specifically less than 89.03%percent89.0389.03\%89.03 %, which significantly trails behind InstGAN. InstGAN demonstrated comparable performance to SOTA diffusion models GDSS and D2L-OMP. All models, including pretraining and tasks involving single properties (QED, logP, SA), as well as multi-property tasks (QED, logP, SA), achieved an overall score surpassing 93.87%. Among GAN-based models, while MolGAN and TransORGAN exhibited a novelty of 100.0%, their validity and uniqueness fell significantly lower than InstGAN. Specifically, MolGAN demonstrated a mere 4.3% uniqueness, and TransORGAN exhibited a validity of only 74.31%. ORGAN’s validity stood at 67.96%, markedly lower than InstGAN’s impressively high validity exceeding 95.45%**B**: Furthermore, InstGAN outperformed SpotGAN in terms of validity, uniqueness, novelty, and total score, with the highest overall performance. InstGAN was trained for single- and multi-property optimization. In single-property optimization, InstGAN was trained individually using QED, logP, and SA. In multi-property optimization, these three properties were used to jointly train InstGAN. The validity, uniqueness, novelty, and total score all reached up to 93.87%percent93.8793.87\%93.87 %. Overall, while InstGAN generated molecules from less informative SMILES strings compared to graph-based models, it surpassed VAE-, flow-, and GAN-based baselines and demonstrated comparable performance to SOTA diffusion models. This showcases InstGAN’s robust capability in molecular generation, excelling in both single- and multi-property optimization. **C**: Table 2 compared the results of InstGAN with various baseline models (including VAE-, flow-, diffusion-, and GAN-based models) for chemical property optimization on the ZINC dataset. To ensure a fair comparison, InstGAN was pretrained several times, and the average results are presented. Additional details on these multiple pretraining sessions are provided in Appendix C. For VAE-based models, although RNN-Attention and TransVAE generated molecules that were highly unique and novel, their validity (i.e., <71.6%absentpercent71.6<71.6\%< 71.6 %) was significantly lower compared to InstGAN. InstGAN outperformed Character-VAE and Grammar-VAE in all metrics. Although JT-VAE exhibited high validity and novelty, the uniqueness was only 19.75%percent19.7519.75\%19.75 %, significantly lower than that of InstGAN

CBA

CAB

BCA

Selection 4

**A**: It may be that one or a few self-supervised tasks are insufficient to learn effective representations, as mentioned in [22]**B**: To further showcase the effectiveness of our method, we compare our algorithm with some recent protein pretraining language models on fold classification [68, 20, 37, 85, 94], which are typically pre-trained with a large scale of data. As shown in Table 6, our method still yields better results without any pre-training or self-supervised learning**C**: This sheds light on the direction of our future efforts: it is interesting to embrace our algorithm with existing protein language models since our core idea is principled.

BCA

BAC

ABC

Selection 3

**A**: Advancing in these two directions constitutes the central motivation of the present work.**B**: The systematic investigation of whether each type of heterogeneity enhances or diminishes the probability of infection has not been systematically undertaken in previous studies**C**: Furthermore, a link between heterogeneous infectivity and microbial growth mechanisms (e.g., microbial rates of birth and death) has yet to be established

BAC

BCA

CAB

Selection 4

**A**: Target novelty (N): The percentage of novel molecules among all the desired valid molecules, the novel molecule is different from training samples;**B**: 3**C**: Target validity (V): The percentage of valid molecules among all the desired molecules, which is measured by RDkit (Landrum et al., 2016) and widely used for calculating validity (Hoogeboom et al., 2022; Xu et al., 2023)); 4

CBA

ABC

CAB

ABC

Selection 3

**A**: Each training image is shown 4 times, and each test image is shown 80 times. We took the average of all trials for each image to form the final dataset. **B**: Each trial in the data is from -0.2 seconds to 0.8 seconds relative to the onset of the stimulus**C**: The training images and test images are presented in separate sessions, but within training and test images all the orders are pseudo-randomized

ABC

CAB

BAC

ABC

Selection 2

**A**: Our drug-ASE matrix analysis (Fig**B**: 6) to data from Google Trends (Fig. 7), we found that the ASEs mentioned most frequently over time are nausea, pain, vomiting, and constipation.**C**: 4) highlights numerous prevalent ASEs based exclusively on social media reports. Additionally, comparing mentions of ASEs on social media in intervals of 14 days (Fig

BCA

ACB

ABC

BCA

Selection 2

**A**: Our approach retrieves drug information using a set of free-to-use application program interfaces (APIs) provided by NLM via representational state transfer (REST) to access the RxNorm and RxTerms datasets: the RxNorm and RxTerms [9], and RxClass [10] APIs**B**: To avoid sending potentially thousands of requests to the NLM servers, we also make use of RxNav-in-a-Box, a locally installable version of the APIs, in combination with Docker [11]**C**: To query the APIs, we created an R [12] package interface which is available for download on GitHub.

BAC

BCA

ABC

Selection 4

**A**: This modification removes sampling error in the fragment generation process, creating an artificially easier learning problem that should result in better performance. We emphasize that ICEBERG-ADV is only suitable for benchmarking: in most realistic C2MS problems ground truth spectra are not available at inference time, meaning that MAGMa cannot be applied.**B**: The stages are trained independently, with the first stage approximating the output of a heuristic fragmenter, MAGMa (Ridder et al., 2014), that uses signals from the spectrum to find a minimal set of fragments that explain the peaks. We introduce ICEBERG-ADV (advantaged), a variant of ICEBERG that replaces the first stage of the model with the exact MAGMa output**C**: ICEBERG (Goldman et al., 2024) is a structured method for high-resolution spectrum prediction that uses a two-stage model. The first stage autoregressively predicts a set of fragments, and the second stage predicts a mapping between those fragments and the peaks in the spectrum (see Appendix C)

CAB

CBA

BAC

ACB

Selection 2

**A**: In addition, we do not cover web-based tools, software and libraries that implement prediction methods (see e.g. [8] for a review). **B**: We limit our focus to new methods that have been tested against large benchmark datasets on pre-clinical synergy and dose response data, in particular leaving out drug-drug interaction (DDI) prediction, which has its own focused literature (see e.g**C**: [7]), as well as papers that are using small scale data or lacking comparison to alternative prediction methods

BCA

ACB

BCA

CAB

Selection 4

**A**: The complexity K⁢(S)𝐾𝑆K(S)italic_K ( italic_S ) measures the number of distinct transformations, including their execution conditions, allowed by the graph transformation model**B**: Such a complexity measure is useful for study and evaluation of empirically inferred input transition systems. In the case of chemical reaction networks, for instance, the complexity allows for formal reasoning about the underlying model of chemistry, opening up possibilities for comparison and further analysis of various industrial and biological processes. **C**: The measure K⁢(S)𝐾𝑆K(S)italic_K ( italic_S ) thus captures the amount of elementary behaviors the model exhibits, or in other words, the diversity of the model dynamics

ACB

BCA

ABC

Selection 1

**A**: These works hypothesise a minimal thermodynamic cost or entropy production per product made, which is related to the information accurately transferred from template to products [2, 31, 32]**B**: However, entropy production in stochastic thermodynamics is related to the relative rate of forwards and backwards transitions [33], rather than the relative rate of two different transitions; it therefore places no minimal cost on the specificity of a catalyst [29]. Exactly how these two ideas are reconciled is unclear. **C**: To the best of our knowledge, only Refs. [2, 31, 32] have considered the implications of two or more pathways for production and destruction of an ensemble of products, in each case without considering a detailed model of dynamical pathways

ACB

CAB

BCA

BAC

Selection 3

**A**: In addition to clustering, the software includes a feature for outlier detection to identify residues that deviate significantly from the expected torsion angle distributions**B**: The software employs a distance-based approach to identify outliers, where residues that fall beyond a certain threshold euclidean distance from the cluster centroids are flagged as outliers with an ’x’ mark. Notably, empirical observations have shown that setting the threshold within the range of 85≤x≤10085𝑥10085\leq x\leq 10085 ≤ italic_x ≤ 100 yields results closely aligned with the validation reports provided by the Protein Data Bank (PDB) for the corresponding models. The outcome of introducing this into the program can be seen in figure 3.**C**: Outliers in the Ramachandran plot may indicate potential issues with the protein model, such as structural anomalies or misalignments

ABC

BCA

ABC

ACB

Selection 4

**A**: (2012). And individuals can be prone to stay away from those with similar opinions corresponding to out-group bias Kimura and Hayakawa (2008). Thus social relationships are under social biases.**B**: (2012); Fu et al**C**: Individuals can voluntarily stay away from those who do not adopt their opinions corresponding to in-group bias Holme and Newman (2006); Durrett et al

BCA

CAB

CBA

BCA

Selection 3

**A**: However, though ABC can potentially be expanded to high-dimensional applications, it is usually only practical to apply ABC when the parameter space is relatively low-dimensional. We propose to focus on inferences for the contagion parameters while considering the true network as a nuisance parameter that is marginalized over. This is most sensible if the purpose of estimation of disease parameters is to parameterize contagion models for the prediction of disease, especially if the model is meant to be generalizable over populations that exhibit varying contact structures but are affected by the same type of disease, such as in the study of outbreaks in two cities with differing population densities. **B**: The unknown contact network can be considered a latent parameter that can be jointly inferred on, alongside the contagion parameters of interest**C**: While MDN-ABC can readily be applied to epidemics on known networks \citepwang2023, it is often necessary to consider the uncertainty introduced by imperfect observations on networks

ACB

BAC

CBA

ABC

Selection 3

**A**: Here, we focused on the widespread standard consolidation theory**B**: However, there exist other models and questions around systems memory consolidation concepts**C**: For instance, the multiple trace theory suggests that some of the hippocampus patterns are conserved in the long term. This theory follows observations of hippocampal damages that produced temporally-graded retrograde amnesia only for semantic memories, but not for episodic ones (Nadel and Moscovitch, 1997).

BAC

ABC

BCA

Selection 3

Datasets:

liangzid
/

robench2024b_all_setq-bioSCP-s

Collection including liangzid/robench2024b_all_setq-bioSCP-s

robench-2024b