The official implementation of AAAI25 paper "Blend the Separated: Mixture of Synergistic Experts for Data-Scarcity Drug-Target Interaction Prediction".
2025/4 📢 News：For easier env configuration, we have separately provided the environment_conda_env_export.yml output by "conda env export" and requirements_pip_freeze.txt output by "pip freeze".
2025/3 📢 News：We have provided the whole dataset and the pretrained kge model (a part of our whole model, which can be reused for every dataset as they use the same KG).
- Download large files. Put drkg.tsv into var_data/ and put kge model into a new folder var_models/. Configure your conda environment according to requirements.txt output by "conda list --export", or the environment_conda_env_export.yml output by "conda env export" and requirements_pip_freeze.txt output by "pip freeze".
- Execute the split_data.py to split the data as cross-validation splits.
- Train and evaluate model from a pretrained KGE model:
You can also train the KGE model yourself without the --load_kge_model argument. You can also save the model components with the --save and load them with the --load* arguments.
python kge/std_main.py --dataset ago_10shots_0 --device 0 --load_kge_model 2024-04-28_10_01_40.24__kgeSLHstd_main.py--save--dataset__a-10--device__6--gate__kge.pth
If there are any issues or cooperation intentions, please contact [email protected].
@inproceedings{zhai2025mosedti,
title={Blend the Separated: Mixture of Synergistic Experts for Data-Scarcity Drug-Target Interaction Prediction},
author={Zhai, Xinlong and Wang, Chunchen and Wang, Ruijia and Kang, Jiazheng and Li, Shujie and Chen, Boyu and Ma, Tengfei and Zhou, Zikai and Yang, Cheng and Shi, Chuan},
booktitle={Association for The Advancement of Artificial Intelligence},
year={2025}
}