Update README.md
Browse files
README.md
CHANGED
@@ -32,14 +32,67 @@ configs:
|
|
32 |
path: "glaser.parquet"
|
33 |
---
|
34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
## Data Sources
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
path: "glaser.parquet"
|
33 |
---
|
34 |
|
35 |
+
# Binding Affinity Dataset
|
36 |
+
|
37 |
+
## Overview
|
38 |
+
|
39 |
+
This dataset is a comprehensive collection of protein-ligand binding affinity data, compiled from multiple sources. The dataset is structured with multiple splits, each corresponding to a specific source:
|
40 |
+
|
41 |
+
- BindingDB IC50 Split
|
42 |
+
- BindingDB Kd Split
|
43 |
+
- BindingDB Kd Filtered Split
|
44 |
+
- BindingDB Ki Split
|
45 |
+
- Davis Split
|
46 |
+
- Davis Filtered Split
|
47 |
+
- KIBA Split
|
48 |
+
- PDBBind 2020 Combined Split
|
49 |
+
- PDBBind 2020 Refined Split
|
50 |
+
- Glaser Split
|
51 |
+
|
52 |
+
In addition to these source-specific splits, a main training split is provided that combines and aggregates data from all these sources.
|
53 |
+
|
54 |
+
## Training Dataset Composition
|
55 |
+
|
56 |
+
The training split is a comprehensive aggregation of multiple molecular binding datasets:
|
57 |
+
|
58 |
+
- Davis-filtered dataset
|
59 |
+
- PDBBind 2020 Combined dataset
|
60 |
+
- BindingDB IC50 dataset
|
61 |
+
- BindingDB Ki dataset
|
62 |
+
- BindingDB Kd Filtered dataset
|
63 |
+
- Glaser dataset
|
64 |
+
|
65 |
+
## Preprocessing Steps
|
66 |
+
|
67 |
+
1. **Dataset Merging**: All specified datasets were combined into a single dataset.
|
68 |
+
2. **Duplicate Removal**: Duplicate entries were dropped to ensure data uniqueness.
|
69 |
+
3. **Binding Affinity Normalization**:
|
70 |
+
- Entries with a binding affinity of 5 were reduced
|
71 |
+
- For duplicate protein-ligand pairs, the mean binding affinity was calculated
|
72 |
+
|
73 |
## Data Sources
|
74 |
+
|
75 |
+
| Dataset | Source | Notes |
|
76 |
+
|---------|--------|-------|
|
77 |
+
| bindingdb_ic50.parquet | [TDC Python Package](https://tdcommons.ai/) | Therapeutic Data Commons |
|
78 |
+
| bindingdb_kd.parquet | [TDC Python Package](https://tdcommons.ai/) | Therapeutic Data Commons |
|
79 |
+
| bindingdb_kd_filtered.parquet | Manually Filtered | See `standardize_data.ipynb` |
|
80 |
+
| bindingdb_ki.parquet | [TDC Python Package](https://tdcommons.ai/) | Therapeutic Data Commons |
|
81 |
+
| davis.parquet | [TDC Python Package](https://tdcommons.ai/) | Therapeutic Data Commons |
|
82 |
+
| davis_filtered.parquet | [Kaggle Dataset](https://www.kaggle.com/datasets/christang0002/davis-and-kiba) | Filtered Davis dataset |
|
83 |
+
| kiba.parquet | [TDC Python Package](https://tdcommons.ai/) | Therapeutic Data Commons |
|
84 |
+
| pdbbind_2020_combined.parquet | [PDBBind](https://www.pdbbind.org.cn/) | Combined PDBBind 2020 dataset |
|
85 |
+
| pdbbind_2020_refined.parquet | [PDBBind](https://www.pdbbind.org.cn/) | Refined PDBBind 2020 dataset |
|
86 |
+
| glaser.parquet | [HuggingFace Dataset](https://huggingface.co/datasets/jglaser/binding_affinity) | Glaser binding affinity dataset |
|
87 |
+
|
88 |
+
## Dataset Columns
|
89 |
+
|
90 |
+
| Column | Description |
|
91 |
+
|--------|-------------|
|
92 |
+
| `seq` | Protein sequence |
|
93 |
+
| `smiles_can` | Canonical SMILES representation of the ligand |
|
94 |
+
| `affinity_uM` | Binding affinity in micromolar (µM) concentration |
|
95 |
+
| `neg_log10_affinityM` | Negative logarithm (base 10) of the affinity in molar concentration |
|
96 |
+
| `affinity_norm` | Normalized binding affinity |
|
97 |
+
| `affinity_mean` | Mean binding affinity for duplicate protein-ligand pairs |
|
98 |
+
| `affinity_std` | Standard deviation of binding affinity for duplicate protein-ligand pairs |
|