This is the repositotry for improving bioactivity predcition using transfer learning.
Goal is to find first tasks/datasets that are more similar to the target task, and train a model on the source task, and then fine-tune the model on the target task.
You can install a package by cloning the repository and running the following command in the root directory of the repository:
pip install -e .
You can use following command to run the code (for training on whole test ChEMBL dataset):
python scripts/training.py --predefined_split --req_training 8 16 32 64 100
You can use following command to run the code (for finetuning on whole test ChEMBL dataset):
python scripts/finetuning.py --predefined_split --req_training 8 16 32 64 100 --k_nearest 10 --strategy selective
--k_nearest is the k nearesr neighbors (training datasets) for the target task. First, we find the k nearest neighbors of the target task, and then we aggregate them and train the model on the aggregated dataset, and then we finetune the model on the target task.
This project is licensed under the MIT License - see the LICENSE file for details
If you use this code in your research, please cite our paper:
@article{fooladi2023fewmol,
title={Machine learning for molecular property prediction: Quantification of task relations and hardness for small organic molecules and proteins},
author={Fooladi, Hosein},
journal={arXiv preprint},
year={2023}
}