Machine learning and graph neural network for finding potential drugs related to multiple myeloma†
Abstract
Drug screening is essential to the process of drug design. The research and development of a new drug requires a lot of time and money. Previous studies have shown that inhibitors of an enhancer of zeste homologue 2 (EZH2) are potential drugs for multiple myeloma (MM). To identify an inhibitor of EZH2, we propose a new virtual screening process, including network pharmacology, a machine learning QSAR model, and a deep learning model. Machine learning and graph neural network methods achieve high performances of 0.83 and 0.26 in terms of R-square and mean square error, respectively. Based on the virtual screening process, a reasonable voting mechanism was established to obtain a comprehensive score of the inhibition ability of small molecule compounds. We successfully screened out drugs for multiple myeloma, formulated with Mulberry leaf and Ganoderma lucidum. Our work may accelerate the process and reduce the cost of drug screening.