A ML code to train and evaluate residue–residue (R–R) contact predictors for proteins.
All experiment settings are written in YAML and are under configs/. Use the --config flag to select a file. Create variants to track experiments (e.g., different loss weights or model heads).
Use main.py as entry point. It accepts the following arguments:
-h, --help show this help message and exit
-d DATASET_PATH, --dataset_path DATASET_PATH
Path to the dataset.
-s SAVING_PATH, --saving_path SAVING_PATH
Path where to save the results.
-c CONFIG_NAME, --config_name CONFIG_NAME
name of the config file in configs directory.
--device DEVICE The device to use (integer) or cpu.
All experiment settings are written in YAML and are under configs/. Use the --config_name flag to select a file within the folder.
Create variants to track experiments (e.g., different loss weights or model heads).
For example, the training I performed for the project can be run using the following commands
python main.py -d ./int_dataset/ -s ./results/proposed_bce -c proposed_bce.yaml
python main.py -d ./int_dataset/ -s ./results/base_bce -c base_bce.yamlContactProject/
├─ configs/ # YAML experiment configs
├─ dataset.py # custom protein dataset & loaders (sequences, contacts, masks)
├─ loss.py # loss functions
├─ main.py # entry point: train/eval
├─ model.py # model wrappers
├─ utils.py # metrics, logging, helpers (collate function)
└─ requirements.txt # Python dependencies
