Spaces:
Runtime error
The Metal3D server does not work?
I am trying to use the metal3D server but the google colab and hugging face both does not work.
Fixed the package import issue due to the upgrade of Python on Colab https://colab.research.google.com/github/lcbc-epfl/metal-site-prediction/blob/main/Metal3D/ColabMetal.ipynb
Huggingface GPU allocation has expired so using Colab will be much quicker.
Colab is working now. Thank you. I was able to download the "probes_model_01.pdb" and "model_01" files and combine them in PyMOL to visualize the protein and the metals. Is there any way to download the metal-interacting residues into the same PDB file or have them in a separate .txt file?
The model does not predict which residues interact with the metal, it only predicts the probability density of zinc since that is what it was trained on. If you use UCSF Chimera you can also easily visualize the cube file and analyze the probability density. The placed probes are indicative but the ground truth for interpretation should always be the probability density (maximum probability is stored in occupancy column of PDB file".
This density can also guide you which residues interact directly with the metal. Based on the typical zinc - protein coordination distance 2.8 A is a good cutoff which you can select in pymol using "model model_01 within 2.8 of model probes_model_01"
Hi,
Super cool tool, thanks for releasing it! Is it possible to look at other metals than zinc? I saw some mention of using the zinc model to predict other metals in the arXiv paper, but is that saying a zinc ion appears where the other metal would be, rather than the accurate metal ion?
Cheersm
Ed
We tested this in our paper Figure 4b https://www.nature.com/articles/s41467-023-37870-6/figures/4
For transition metals it is very hard to distinguish the binding sites and most of the zinc binding sites will also bind copper or other transition metals since ionic radii etc are pretty similar.
So while technically the model since it was trained only on zinc only predicts a density map for zinc binding in practice it will give you good results for most transition metals.
But it won't tell you which metal it is.