Pin-Tuning: Parameter-Efficient In-Context Tuning for Few-shot Molecular Property Prediction

Implementation for paper: Pin-Tuning: Parameter-Efficient In-Context Tuning for Few-shot Molecular Property Prediction

This is the code for the NeurIP'24 Paper: Pin-Tuning: Parameter-Efficient In-Context Tuning for Few-shot Molecular Property Prediction.

Usage

For data used in the experiments, please download data.zip from this page, then extract the downloaded file and save the contents in the ./data/ directory.

For quick start, you could run the scripts in the ./scripts/ directory using the following command. Make sure to modify the GPU ID in the scripts according to your actual setup.

sh scripts/10shot/run_<dataset_name>.sh # for 10-shot setting
# example: sh scripts/10shot/run_tox21.sh

sh scripts/5shot/run_<dataset_name>.sh # for 5-shot setting
# example: sh scripts/5shot/run_sider.sh

Supported datasets:

• tox21, sider, muv, pcba
• toxcast-APR,toxcast-ATG,toxcast-BSK, toxcast-CEETOX,toxcast-CLD,toxcast-NVS, toxcast-OT,toxcast-TOX21,toxcast-Tanguay

Requirements

• Python >= 3.9
• PyTorch >= 1.12.1
• torch_geometric >= 2.3.1
• torch_scatter==2.1.0
• rdkit==2023.3.3
• learn2learn==0.2.0
• numpy==1.26.4
• scikit_learn==1.4.0
• seaborn==0.13.2
• tqdm==4.66.1

Citation

Please cite our paper if you use the code:

@inproceedings{wang2024pintuning,
  author       = {Liang Wang and Qiang Liu and Shaozhen Liu and Xin Sun and Shu Wu and Liang Wang},
  title        = {Pin-Tuning: Parameter-Efficient In-Context Tuning for Few-Shot Molecular Property Prediction},
  booktitle    = {NeurIPS},
  year         = {2024}
}