HIV-1 Drug Efficacy Identification via Informative Knowledge Graph Selection on Graph Neural Networks

Human immunodeficiency v irus t ype-1 (HIV-1) cases have recently grown, remaining a worldwide problem. The only action to deal with HIV-1 disease is to take the drug resistance or HIVDR. The formulation of HIVDR is based on antiretroviral therapy (ART) personalization until we know its efficacy. I n t his w ork, w e a im t o a ccurately a ssess t he efficacy of HIV drugs (HIVDE) by learning from informative clinical and treatment data alone. Hence, we propose an integration method between a metaheuristic algorithm for selectively choosing important clinical and treatment features and graph neural networks for HIVDR identification. Once the important features are in hand, we attempt to draw their contextual connectivities or knowledge graphs. After careful design, the proposed method is a combination of a harmony search algorithm for feature selection, a graph encoder with angular distance for knowledge graph (KG) construction, and graph convolutional networks for the HIVDE identification. Since this work application is limited, we evaluated the proposed method on a particular public HIVDR dataset. Interestingly, the”patient response” feature has shown no effect in identifying the efficacy of HIV drugs. Furthermore, the feature selection and the KG construction succeed in boosting the identification, which also outperforms the baseline methods.