PhyIndBC: Development of a machine learning tool for screening of potential breast cancer inhibitors from phytochemicals

Abstract

Breast cancer is the foremost contributor to cancer-related mortality among women on a global scale. However, its treatment encounters challenges compounded by the disease's complexity. A promising avenue in the quest for effective therapeutics lies within the realm of phytomolecules, which are characterized by their chemical diversity and biological potential. Thus, in the current study a machine learning (ML) model was created using phytomolecules having inhibitory activity against breast cancer cell lines. Multiple ML techniques viz., k-nearest neighbor (KNN), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB) were combined with various molecular fingerprints (MACCS and Morgan2) to develop multiple predictive models. Among these models, the RF algorithm coupled with the MACCS fingerprint emerged as the best performing model. Mean decreases in impurity, t-SNE analysis, and k-means clustering was studied to determine the important features and understand chemical space diversity. Further, to predict potential breast cancer inhibitors, ADMET adherent Natural Products (NPs) of plant origin (identified from the COCONUT database) were screened using the developed ML model. NPs predicted as actives were further screened via ensemble virtual screening (eVS) technique against erb-b2 receptor tyrosine kinase 2 (HER2), to identify high-affinity molecules against this breast cancer drug target. In summary, the validated machine learning model developed in this study has been incorporated into a freely available standalone package named PhyIndBC (https://github.com/subhashmagarwal/PhyIndBC) which can be used for virtual screening and predicting breast cancer inhibitors of plant origin.