Comparative Evaluation of Explicit Solvent Models for RNA-Ligand Docking.

The interest in targeting RNA with small molecules is increasing continuously. However, structure-based drug design approaches have been reported rarely so far. Major challenges in RNA-ligand docking include ligand-induced conformational changes, ions and solvation which hamper successful applications in prospective virtual screenings. We examined the influence of explicit solvent inclusion on RNA-ligand docking performance using crystallographic water sites as well as the computational solvation models 3D-RISM, GalaxyWater-CNN and waterdock_fxx in combination with FlexX, FlexX with HYDE rescoring, GOLD and LeadIT docking. The redocking study with 92 RNA-ligand complexes underlined that the benefit of solvent consideration is highly target-specific and resolution-dependent reaching on average accurate pose predictions of around 70% for all structures and only 35% for low-resolution structures for FlexX, GOLD and LeadIT. HYDE performed slightly worse on average with an overall 50% accurate pose prediction and varying impact of predicted solvent. Success rates of structures lacking experimental solvent information were improved by involving predicted water sites. 3D-RISM predictions showed most robust results across all resolutions, improving success rates by up to 30% for low-resolution structures in combination with LeadIT. In addition, NMR and ion-free structures were found to be more challenging in pose prediction accuracy compared to ion-containing X-ray structures. Cross-docking studies across five representative RNA targets demonstrated improvements for hydrated dockings, while different binding site conformations indicated RNA dynamics as an additional challenge. The best cross-docking setup was partially deducible from the corresponding redocking setup revealing great potential to advance virtual screenings by the inclusion of explicit solvent sites.