The ensemble framework of EnsemPPIS for predicting protein-protein interaction sites (PPIS):

Please follow these steps to train the EnsemPPIS for the prediction of PPIS:

1. Create a Python virtual environment (version=3.7.11) according to the 'requirement.txt' file;

2. Put your sequence data into the 'data_cache' folder, and run 'data_preprocess.py' in this folder to generate three pkl files, namely, encode_data.pkl, label.pkl and dset_list.pkl;

3. Download the pre-trained ProtBERT model (pytorch_model.bin) from http://ensemppis.idrblab.cn/download_ProtBERT, and put it into the 'feature_generator' folder;

4. Run 'ProtBERT_feature_generator.py' in the 'feature_generator' folder to generate ProtBERT enbeddings for sequences;

5. Run 'main-TransformerPPIS.py' to train the TransformerPPIS model;

6. Run 'main-GatCNNPPIS.py' to train the GatCNNPPIS model;

Predict PPIS using the trained EnsemPPIS:

7. Run 'predict_EnsemPPIS.py' to predicte PPIS using the trained TransformerPPIS and GatCNNPPIS.

Datasets:

All the benchmark datasets used in this study was provided in the 'datasets' folder.

Training commond:

The two base models of EnsemPPIS (TransformerPPIS and GatCNNPPIS) were trained separately.

The commond for training TransformerPPIS on CPU:

nohup python main-TransformerPPIS.py > out-TransformerPPIS.txt 2>&1 &

The commond for training TransformerPPIS using single GPU or distributed training using multiple GPUs: single GPU:

CUDA_VISIBLE_DEVICES=1 nohup python -m torch.distributed.launch --nproc_per_node=1 --master_port 6666 main-TransformerPPIS.py > out-TransformerPPIS.txt 2>&1 &

multiple GPUs:

CUDA_VISIBLE_DEVICES=1,2 nohup python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 main-TransformerPPIS.py > out-TransformerPPIS.txt 2>&1 &