Updates

Full Data: Google Drive

---

Data

The data are available in the 'dataset' folder. Due to file size limitations, only the drug indexes for each fold and a small sample of data are provided to ensure reproducibility. The full dataset and model weights will be released on GitHub at a later time. "model_name_or_path": "output/mult_task_training_base_drugbank_transductive_5e-4_fold2/runtime/checkpoint-24540",

This GitHub repository excludes large raw datasets, generated caches, local Conda environments, and model checkpoints. Use data/dti_template/sample_dti.json for the lightweight runtime smoke test, and download full data/model artifacts from the Google Drive link above when needed.

Code

For all the training and inference scripts, first modify the configs file in the 'config' folder, then run 'python xxx.py' from the root folder.

Prediction model:

'finetune_molt5_prediction.py' is used for training and inference.

ExDDI-S2S:

'finetune_molt5_joint_training.py' is used for training and inference.

ExDDI-MT:

'finetune_molt5_mult_task.py' is used for training and inference.

ExDDI-MTS:

'finetune_molt5_mult_task_inference.py' is used for inference. (Models are trained by ExDDI-MT).