Code to reproduce the experiments from the paper Systematically Exploring QSAR Models for Activity-Cliff Prediction. This repository also contains clean molecule- and MMP data for all three analysed data sets (dopamine receptor D2, factor Xa, SARS-CoV-2 main protease) as well as the original numerical results from the experiments conducted in the paper.
The data-folder contains three clean chemical data sets of small-molecule inhibitors of dopamine receptor D2, factor Xa, or SARS-CoV-2 main protease respectively. Each data set is represented by two files: molecule_data_clean.csv and MMP_data_clean.csv. The first file contains SMILES strings with associated activity values and the second file contains all matched molecular pairs (MMPs) identified within the first file.
The experiments in the paper can be reproduced by running the code in the Jupyter notebook QSAR_activity_cliff_experiments.ipynb. First, the QSAR-, AC-, and PD-prediction tasks for the chosen data set are formally constructed in a data-preparation section. Then, an appropriate data split is conducted, both at the level of individual molecules and MMPs. Finally, a molecular representation (PDV, ECFP, or GIN) and a regression technique (RF, kNN, MLP) are chosen and the resulting model is trained and evaluated for QSAR-prediction, AC-classification and PD-classification. The computational environment in which the original results were conducted can be found in environment.yml.
The experimental results can be visually explored using the visualise_results-function at the end of QSAR_activity_cliff_experiments.ipynb. This function produces scatterplots such as the one in the graphical abstract above. The original numerical results from the paper are saved in the resuls-folder; thus the original plots from the paper (and more) can be generated with visualise_results.