PCANN - Protein Complex Affinity Neural Network

This repository contains the scripts, trained models and sample data to predict the binding free energy of a protein-protein complex based on its 3D-coordinates. The package includes 25 models, whose results are averaged to obtain the prediction for $$∆G^{bind}$$

System requirements

Key packages and programs:

• Linux platform (tested on ubuntu 20.04 and ubuntu 22.04)
• python3 (tested with python3.9)

Installation dependencies

Install all dependencies except for PyTorch and PyTorch Geometric.

# create virtual enviroment
python3.9 -m venv ./venv
source ./venv/bin/activate
pip install --upgrade pip
pip install -U setuptools wheel pip

# install python packages
pip install -r requirements.txt

# install PCANN in PYTHONPATH
python setup.py install

# download pretrained esm model
cd trained_models/ESM-2/
bash download.sh

Note! Due to the variations in GPU types and drivers, we are unable to make one environment that will run on all systems. As such, we provide only the file requirements.txt with support for CUDA 11.3, leaving it to a user to customize this file. The customization involves changing the cudatoolkit, PyTorch version and python version in requirements.txt

Make predictions. Example

We provide the templates to make predictions in examples/make_predictions/ Please, change path_to_dimer_pdb parameter in examples/make_predictions/load.py to point at the PDB structure (protein-protein complex) of interest

   # run the script to make predictions
   cd examples/make_predictions/
   python load.py

See the expected result in examples_output/make_predictions/ Predictions from the 25 PCANN models along with the averaged result can be found in examples_out/make_predictions/output/output.csv. We recommended to use the PCANN(avg) scheme

Run training and testing. Example

In a way of a demo, we provide the template to train and test the PCANN model

   # run the script to train and test PCANN 
   cd examples_out/run_training
   python run_train.py
   python run_training/run_test.py