Harnessing machine learning models to repurpose drugs targeting HIV-1 integrase, protease, and reverse transcriptase.

The human immunodeficiency virus type 1 (HIV-1) is a type of retrovirus responsible for immunodeficiency syndrome, a sickness that significantly compromises the immune system's ability to function. The latest data on the global HIV epidemic reveals that 39,9 million people are living with HIV, 1,3 million have newly acquired the virus, and 630,000 have died from HIV-related illnesses. Given the high costs and complexities of experimental screenings, this study proposes an ensemble approach that combines machine learning and molecular docking to rank potential HIV-1 drug candidates for repurposing. This integrative approach targeting multiple key HIV-1 enzymes (protease, integrase, and reverse transcriptase) contributes to ongoing efforts to identify and repurpose potential inhibitors, completing existing studies focused on one or two molecular targets. The meta-learning based integrative strategy demonstrated robust predictive performance across all three HIV-1 targets. For HIV-1 integrase, the meta-learning model mllh-18-hp achieved a Matthews correlation coefficient (MCC) of 99.14% and a precision-recall area under the curve (PR-AUC) of 99.68%, while the mllh-18 model reached 98.28% MCC and 99.36% PR-AUC. For HIV-1 protease, the mllh-18 and mllh-18-hp models yielded MCC values of 97.20% and 95.29%, with PR-AUCs of 99.96% and 99.97%, respectively. In the case of HIV-1 reverse transcriptase, the mllh-18-hp model recorded 97.49% MCC and 99.61% PR-AUC, whereas the mllh-18 model showed 96.07% MCC and 99.32% PR-AUC. These models were subsequently used to screen the DrugCentral database, and compounds classified as active were docked and integrated into a pipeline to prioritize the most promising candidates for repurposing. The results indicate that three drugs (enoxacin, larotrectinib, and pipamazine), originally developed for other diseases, have potential as candidates for HIV-1 management, but experimental tests are advised. Beyond the explored potential for repurposing HIV-1 drugs, this research highlights an meta-learning based approach that could improve the prediction of potential candidates and provide useful insights for future drug development efforts.