Metal3D and Metal1D: Accurate prediction of transition metal ion location via deep learning

If using this work please cite:

Accurate prediction of transition metal ion location via deep learning S.L. Dürr, A. Levy, U. Rothlisberger bioRxiv 2022.08.22.504853; doi: https://doi.org/10.1101/2022.08.22.504853

How to run predictions

No installation, no account required use Metal3D on Huggingface Spaces

If you prefer a notebook based environment

Command line usage is described below

How to install and run locally

For local installation run the following commands to setup the environment.

conda env create -f environment.yml
conda activate metalprediction
cd Metal3D

You need to have VMD installed to view predictions directly from the commandline program (connect with ssh -X if working on a remote machine), download VMD from uiuc.edu. Alternatively you can use --writecube --cubefile nameofcube.cube --softexit and view the predicted maps in UCSF Chimera or any other viewer that supports cube file.

Typical commands would be:

Analyze all ASP, CYS, ASN, GLN, GLU and HIS residues, write a pdb file with the found probes and write the maximum probabilty to a text file. Will open VMD viewer.

./metal3d.py --pdb PDB.pdb --metalbinding --writeprobes --probefile metalsites.pdb --maxp

Analyze only specific residues in the pdb file and write a cubefile to disk without openening VMD. ./metal3d.py --pdb PDB.pdb --id 91 94 116 --writecube --cubefile test.cube --softext

Display all possible options ./metal3d.py --help

Data

The PDB codes used for training, validation and testing are available in data. The PDB codes used for the selectivity analysis including the residue ids of the coordinating residues are available in data as selectivity_analysis_sites.csv.

License

All code is licensed under MIT license, the weights of the network are licensed under CC BY 4.0.