AiPPA

Description

AiPPA is a GNN model for protein-protein binding free energy prediction without the need for the protein complex structures.

Dependencies

AiPPA was built and tested on Python 3.9 with PyG 2.3. To install all dependencies, directly run:

cd    AiPPA-main
conda env create -f environment.yml
conda activate AiPPA

Affinity Prediction

Quick start

Use AffinityPrediction.py to predict the binding free energy between any two proteins.

python AffinityPrediction.py -pA proteinA.pdb -pB proteinB.pdb

Prediction on a dataset

1. Data Preparation

The example data files are located in the data folder. Please prepare the necessary data for training or testing. For convenience, the data folder can be organized as follows.

.
├── affinity.txt
└── pdbs
    ├── ComplexName1
    │   ├── ComplexName1_proteinA.pdb
    │   └── ComplexName1_proteinB.pdb
    └── ComplexName2
        ├── ComplexName2_proteinA.pdb
        └── ComplexName2_proteinB.pdb

2. pdb2graph

Navigate to the features folder, and run generateGraph.py to convert protein structure files into graphs. The resulting pickle files will be saved in features/pkls:

python generateGraph --data_dir ../data/pdbs \
                     --target_path ./pkls \
                     --target_file ../data/affinity.txt

3. Train or Test

Run Runfile.py to train.

Run BenchmarkTest.py to test on your own dataset.

MonteCarloSampling

MonteCarloSampling contains the code for de novo nanobody design that integrates AiPPA with thermodynamic Monte Carlo sampling, as described in our submitted manuscript: "De novo nanobody design using graph neural networks and thermodynamic Monte Carlo sampling."

Authors and article to be cited

• Lei Wang, Xiaoming He, Gaoxing Guo, Xinzhou Qian, and Qiang Huang*. De novo nanobody design using graph neural networks and thermodynamic Monte Carlo sampling. (Submitted)