GenZProt : Chemically Transferable Generative Backmapping of Coarse-Grained Proteins

This is a repo accompanying our paper "Chemically Transferable Generative Backmapping of Coarse-Grained Proteins" (arxiv link).

We propose a geometric generative model for backmapping all-atom coordinates from (coarse-grained) C_alpha traces of proteins.

Download PED data

cd data
wget https://zenodo.org/record/7683192/files/genzprot_pedfiles.tar.gz
tar -xvf genzprot_pedfiles.tar.gz

Install packages

You can install the requirements via conda. pytorch=1.11.0=py3.8_cuda11.3_cudnn8.2.0_0

conda env create -f environment.yml

Download and install this package

git clone https://github.com/learningmatter-mit/GenZProt
cd GenZProt
conda activate genzprot
pip install -e . 

Backmapping C_alpha traces into all-atom structures

Saved checkpoint of GenZProt is located in './ckpt/'. Save your C_alpha traces in .pdb format and pass its path to 'ca_trace_path' argument. We need an all-atom pdb file (at least one model/frame) to get the topology and C_alpha mapping. Pass the path to 'topology_path' argument. Note that GenZProt does not support the backmapping of PTMs, water, or HETATMs. Please remove all PTM atoms (e.g., OXT) and HETATMs from the all-atom pdb file before running the code.

Example:

cd scripts
MPATH=../ckpt/model_seed_12345
ca_trace_name=PED00055_CA_trace
ca_trace_path=../data/${ca_trace_name}.pdb
top_path=../data/PED00055.pdb
python inference.py -load_model_path $MPATH -ca_trace_path $ca_trace_path -topology_path $top_path

The results are saved in both .npy ( shape = ( 10,n_cg_samples,n_atoms_truncated,3 ) ) and .pdb format, in a directory named result_{MPATH}_{ca_trace_name}. Because our algorithm requires i-1th and i+1th C_alpha positions to locate the atoms of the ith residue, it does not backmap the first and the last residue. Hence, n_atom_truncated = n_atom - (n_atom_first_res + n_atom_last_res).

Training your own GenZProt

cd script
python train_model.py -load_json modelparams/multi.json

Test script

cd script
MPATH=../ckpt/model_seed_12345
test_data_path=../data/PED00055e000.pdb
python test_model.py -load_model_path $MPATH -test_data_path $test_data_path