APTAnet

An atom-level PTI affinity prediction model

1. environment configuration

conda env create -f environment.yaml
torch.cuda.is_available()

2. Code Structure

• The "data" folder contains the data, which is primarily sourced from https://github.com/PaccMann/TITAN and various databases.
• The "experiment" folder includes experiment results files and parameter files.
• "pretrain.py" is the model training code.
• "mydataset.py" is responsible for data loading.
• "APTAnet.py" defines the APTAnet model.
• "knn.py" is the KNN baseline model.
• "plot" contains the code for generating plots.

3. Usage

• Activate Conda environment

• check experiment/model_params.json

• In the command-line interface, initiate DDP (Distributed Data Parallel) training, as shown in the example below, using two GPUs:

  CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.run --nnodes=1 --nproc_per_node=2 --node_rank=0  pretrain.py
  

• The training results can be found in the "experiment" folder.