ProMIM - Multi-level Interaction Modeling for Protein Mutational Effect Prediction

Official repository for "Multi-level Interaction Modeling for Protein Mutational Effect Prediction".

Figure: The overview framework of ProMIM.

Multi-level Interaction Modeling for Protein Mutational Effect Prediction
Yuanle Mo*, Xin Hong*, Bowen Gao, Yinjun Jia, Yanyan Lan

Description

Protein-protein interactions are central mediators in many biological processes. Accurately predicting the effects of mutations on interactions is crucial for guiding the modulation of these interactions, thereby playing a significant role in therapeutic development and drug discovery. Mutations generally affect interactions hierarchically across three levels: mutated residues exhibit different sidechain conformations, which lead to changes in the backbone conformation, eventually affecting the binding affinity between proteins. However, existing methods typically focus only on sidechain-level interaction modeling, resulting in suboptimal predictions. In this work, we propose a self-supervised multi-level pre-training framework, ProMIM, to fully capture all three levels of interactions with welldesigned pre-training objectives. Experiments show ProMIM outperforms all the baselines on the standard benchmark, especially on mutations where significant changes in backbone conformations may occur. In addition, leading results from zero-shot evaluations for SARS-CoV-2 mutational effect prediction and antibody optimization underscore the potential of ProMIM as a powerful next-generation tool for developing novel therapeutic approaches and new drugs.

If you find this code useful, please consider to star this repo and cite us:

@article{mo2024multi,
  title={Multi-level Interaction Modeling for Protein Mutational Effect Prediction},
  author={Mo, Yuanle and Hong, Xin and Gao, Bowen and Jia, Yinjun and Lan, Yanyan},
  journal={arXiv preprint arXiv:2405.17802},
  year={2024}
}

Environment Setups

1. Clone the Repository

git clone https://github.com/mylRalph/ProMIM.git
cd ProMIM
git clone https://github.com/anton-bushuiev/PPIRef.git

2. Prepare the Environment

For the promim environment, we suggest installing the unicore package from Uni-Core. The ppiref environment is used for parsing the PPIRef50K dataset. Please refer to the official PPIRef repo for detailed instructions.

conda create -n promim python=3.8
conda activate promim
pip install -r requirements.txt
conda create -n ppiref python=3.10

Datasets

1. Get PPIRef50K

Download PPIRef50K dataset, and split it into training and validation sets by running:

conda activate ppiref
python ./data/get_ppiref50k.py

Use the following command to preprocess PPIRef50K:

conda activate promim
python ./src/datasets/ppiref50k.py

2. Get SKEMPI2

cd data
bash ./get_skempi_v2.sh
cd ..

Inference

Download the trained weights from Google Driver and put them into the trained_models folder. We use the trained weight of RDE from the RDE-PPI repo for SIM.

1. Predicting Mutational Effects on Protein-Protein Binding

bash ./scripts/test/test_promim_skempi.sh

Remember to specify the idx_cvfolds parameter to evaluate on the particular fold. You can chose from 0,1,2.

ckpt=./trained_models/promim_skempi_cvfold_2.pt
device=cuda:0
idx_cvfolds=2

python test_promim_skempi.py \
    --ckpt $ckpt \
    --device $device \
    --idx_cvfolds $idx_cvfolds

2. Predicting Mutational Effects on Binding Affinity of SARS-CoV-2 RBD

bash ./scripts/test/test_promim_6m0j.sh

3. Optimization of Human Antibodies against SARS-CoV-2

bash ./scripts/test/test_promim_7fae.sh

Training

1. Train ProMIM

bash ./scripts/train/train_promim.sh

You can set the wandb flag and wandb_entity parameter in train_promim.sh to use Weights & Biases for logging or use TensorBoard by default.

nproc_per_node=4
world_size=4
master_port=20888
config_path=./configs/train/promim.yml

CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=$nproc_per_node --master_port=$master_port train_promim.py \
    --config $config_path \
    --world_size $world_size \
    --wandb \
    --wandb_entity your_wandb_username

2. Train ProMIM DDG Predictor

bash ./scripts/train/train_promim_skempi.sh

Cite us

If you find our work useful, please cite our paper:

@article{mo2024multi,
  title={Multi-level Interaction Modeling for Protein Mutational Effect Prediction},
  author={Mo, Yuanle and Hong, Xin and Gao, Bowen and Jia, Yinjun and Lan, Yanyan},
  journal={arXiv preprint arXiv:2405.17802},
  year={2024}
}

License

The code is released under the MIT license.

THU-ATOM/ProMIM