Deep learning challenge to generate apo RNA conformations with cryptic binding site

RNA plays vital roles in diverse biological processes, thus drawing much attention as potential therapeutic target. Notably, cryptic ligand binding sites can be promising targets due to high specificity in RNA-ligand interaction. Structure Based Drug Design (SBDD) for RNA is a practical approach to achieve successful drug candidates. However, RNA-targeted SBDD faces a technical challenge, a paucity of available tertiary structures of RNA-ligand complexes. To address the problem, we present a computational strategy to generate RNA conformations harboring cryptic binding sites from apo RNA conformations, by using Molearn, a generative deep learning (DL) model designed to explore high energy states along conformational transition pathways. Focusing on the paradigmatic RNA-ligand complex, HIV-1 Transactivation Response Element (TAR) binding to MV2003, we succeed to generate TAR conformations with a cryptic binding cavity for MV2003, without a priori knowledge of the cavity. These conformations were docked to MV2003, yielding poses with binding scores comparable to those calculated for experimentally resolved TAR-MV2003 complexes. Prior atomistic molecular dynamics and DL-based RNA 3D structure prediction studies on this target have not reported recovery of the cryptic site from apo conformations. Finally, we discuss the technical advantages, acknowledge current limitations, and outline concrete avenues for improvement. To our knowledge, this is the first report of theoretical generation of RNA conformations with a cryptic ligand binding site. This finding demonstrates a baseline for future studies in this direction and suggests a practical route to advance RNA-targeted SBDD.