MuToN: An end-to-end geometric deep learning framework for predicting binding affinity changes upon protein mutations.

Requirements

• BioPython* (v1.78). To deal with computings relating to structures and sequences of proteins.
• Pytorch* (v2.0.1). Pytorch with GPU version. Use to model, train, and evaluate the actual neural networks.
• Scikit-learn* (v0.24.1). For machine learning relating computation.
• ESM*. For protein residue embedding.
• Modeller*. For mutant protein structure modeling.
  Installing refers to https://salilab.org/modeller/download_installation.html.
  You need a license key to use Modeller first.
  An easy way is to install Modeller using the 'conda' package manager, simply run from a command line:
  conda config --add channels salilab
  conda install modeller
  And then Edit the file config.py according to promption after installation by replacing XXXX with your Modeller license key.

Download data.

We open sourced all the precomputed data using to reproduce the results of our paper https://zenodo.org/records/10445253.
This repository includes the lists of mutation records, complexes, mutant protein structures and files of pre-computed LLM embeddings.

wget https://zenodo.org/records/10445253/files/data.zip 
unzip data.zip

Training and evaluation.

The project does not contain a standalone script specifically for preprocessing protein structures. However, during the data loading stage, the protein preprocessing routine is executed.

Usage description:
--checkpoints_dir. #Default is Checkpoints/example. Specify a directory to save the model checkpoints.
--dataset SKEMPI2 or S1131 or S4169 or M1101. #Default is S1131. Specify which dataset to use.
--splitting mutation or complex. #Default is mutation. Specify the Train-Test partitioning mode of the dataset. Mutation-level or complex-level.
--device cuda:0 or cpu, etc. #Default is cuda:0. Specifies the device to run the model on.
For example, to train and evaluate the model on the SKEMPI2 dataset using the mutation-level splitting mode, run the following command:

python train.py --checkpoints_dir SKEMPI2_mutation --dataset SKEMPI2 --splitting mutation --device cuda:0

License

MuToN is released under an MIT License.

MuToN quantifies binding affinity changes upon protein mutations by geometric deep learning.