Application of machine learning in virtual screening of HIV integrase inhibitors

Abstract: Applications of machine learning in drug design is an emerging and fast-growing field of research. In silico models allow the speeding up of drug discovery and developments. HIV, once known as “the disease of the century”, is currently with no specific treatment. New drugs discovered by machine learning may have the potential to complement ARV-therapy to help prolong the lifespan of people who live with HIV. The machine learning model uses the multi-layer perceptron (MLP) architecture along with the Extra Trees algorithm for feature selection, and handling imbalanced data with the Tomek Links algorithm gave the results: 95.3% accuracy, 82.6% sensitivity, 86.5% precision, F1 score of 0.845, area under the ROC curve (0.953) and Average Precision (0.917). The internal dataset was screened through the PAINS filter and QSAR machine learning models and substance DI081 was determined to be the best candidate with the probability of inhibiting HIV integrase up to 98.57%. Download paper here

Recommended citation: Tieu Long Phan, Xuan Truc Tran Dinh, The Chuong Trinh, Hoang Son Le Lai, Ngoc Tuyen Truong (2022). Application of machine learning in virtual screening of HIV integrase inhibitors - Vietnamese Pharmacy Journal, 506, 21-25