RLSFmode: A deep learning approach for predicting RNA-small molecule binding modes via molecular surface modeling

As RNA becomes an emerging therapeutic target, predicting the binding mode of RNA-small molecule through emerging algorithms is of crucial importance in drug development. A complete prediction of binding modes should not only include recognition of native poses but also estimate binding affinity for a given structure. Accurately identifying binding poses can reveal specific interaction patterns between small molecules and RNA, providing a foundation for structure based rational drug design. High precision prediction combined with affinity can help efficiently screen candidate molecules with potential therapeutic functions, significantly reducing experimental costs and development cycles. However, due to the complexity of RNA binding pockets and the scarcity of structural data, the development of existing computational models has encountered significant challenges. In this study, we developed a deep learning model RLSFmode via molecular surface modeling, which completed the binding mode prediction task by combining RNA-ligand surface fingerprinting with sequence information and energy scoring functions. After rigorous testing, RLSFmode has demonstrated significantly improved performance.