Abstract
Human immunodeficiency virus (HIV) integrase inhibitors play a critical role in antiretroviral therapy, but the emergence of drug resistance necessitates the discovery of novel compounds. Machine learning (ML) offers a data-driven approach to accelerate drug discovery by predicting potential inhibitors with high efficacy. This study utilized a curated dataset of known HIV integrase inhibitors and employed feature engineering techniques to extract molecular descriptors. Random forest and logistic regression models were trained to classify compounds based on their inhibitory potential. Model performance was evaluated using accuracy, precision, recall, and the area under the receiver operating characteristic curve (AUC-ROC). The random forest model demonstrated superior predictive performance, achieving an AUC-ROC of 0.886, an accuracy of 0.815, and a precision of 0.79. Key molecular features, including hydrogen bond donors, rotatable bonds, and molecular weight, were identified as crucial determinants of inhibition. The models successfully screened novel compounds with high predicted inhibitory potential. Machine learning (ML) provides a powerful tool for the rapid identification of potential HIV integrase inhibitors. This study highlights the importance of molecular descriptors in predicting inhibitory activity and demonstrates the feasibility of ML-driven drug discovery. Future work will focus on refining model generalization, expanding datasets, and developing a user-friendly platform via Streamlit to enhance accessibility for researchers and drug developers.