[**Paper Link**](https://pubs.acs.org/doi/pdf/10.1021/acs.jcim.2c01564)

[**Supporting Information**](https://pubs.acs.org/doi/suppl/10.1021/acs.jcim.2c01564/suppl_file/ci2c01564_si_001.pdf)

## Keyword
- RF, GB, SVM, XGBoost
- GCN, GAT, Attentive FP
- Transfer Learning
- Shapley additive explanations (SHAP) method
- matched molecular pair analysis (MMPA)

## Transfer Learning
- Pre-trained on computational data
- Fine-tuned on target dataset

Other examples of pre-training
- ChemNet
- MGBert
- K-Bert

## Hyerparameter Optimization
- Grid Search: RF, GB, SVM, XGBoost
- Tree-Structured Parzen Estimators (TPE): GCN, GAT, Attentive FP
- Early Stopping (if not improved in 10 successive epochs)

## Explainability
- SHAP (for XGBoost)
- Attention Mechanism (for GAT)
- MMPA

## Reference
- [ChemNet](https://arxiv.org/pdf/1712.02734.pdf)
- [MGBert](https://pubmed.ncbi.nlm.nih.gov/33951729/)
- [K-BERT](https://academic.oup.com/bib/article/23/3/bbac131/6570013)
- [SHAP](https://arxiv.org/pdf/1905.04610.pdf)
- [MMPA](https://www.knime.com/knime-analytics-platform)
- [MMPA in RDKit](https://zhuanlan.zhihu.com/p/389763022)