transformerCPI_DTA

Refine https://github.com/lifanchen-simm/transformerCPI/ for large dataset and multiple GPU training

Install Requirements

conda env create -f py36_tCPI.yml

Training and testing

First run sh script/generate_map.sh to generate protein_map.pkl and smiles_map.pkl, which is the mapping from smiles to smiles_feature, and protein_seq to protein_seq_feature

Then run sh script/main.sh to start training.

If you want to stop the training process, run sh script/stop.sh

Comparison to the orginal repo https://github.com/lifanchen-simm/transformerCPI

Advantages:

1. You can use torch.nn.DataParallel(along with torch.cuda.amp) to accelerate your training process.
2. For large scale dataset, Using DTADataset in DataUtil.py along with torch.nn.DataLoader can accelerate your data loading process and tremendously reduce the memory usage.
3. I change the code into solving regression problem instead of classification problem in the original paper.

Thanks for the brilliant work of authors in this paper https://doi.org/10.1093/bioinformatics/btaa524