AutoML in Drug Discovery: Side-Effects Prediction Using AutoGluon Framework and Its Applications in Drug Discovery

We present herein an efficient AutoML based approach using AutoGluon framework for the prediction of 1385 side-effects for the first time in literature, which offers significant advantages over reported approaches, in terms of improved predictive performance and also, saving time and thus, cost of drug discovery. In the present study, we employed i) 1385 side-effects data obtained from SIDER database for 888 drugs. ii) AutoGluon (AG). iii) three machine learning (ML) techniques - Support Vector Machine (SVM), Random Forest (RF), Artificial Neural Network (ANN). iv) Sparse Canonical Correlation Analysis (SCCA) method. v) 881 PubChem fingerprints chemical features from PaDEL software. AutoGluon outperformed all the other manually executed ML methods and published results.