Can Current Molecular Docking Methods Accurately Predict RNA Inhibitors?

Area of Science:

• Computational biology
• Drug discovery
• Molecular modeling

Background:

• Noncoding RNAs are significant in cancer and represent promising drug targets.
• Conventional drug screening is resource-intensive, making computational methods like RNA-ligand docking a viable alternative.
• Existing docking tools are primarily optimized for protein-ligand and protein-protein interactions, not RNA-ligand systems.

Purpose of the Study:

• To evaluate the performance of commonly used docking software (AutoDock Vina, HADDOCK, HDOCK) and a specialized RNA-ligand tool (RLDOCK).
• To assess these methods based on cognate docking, blind docking, scoring, and ranking potentials for RNA-ligand interactions.
• To identify the limitations of current docking approaches in accurately predicting RNA-ligand binding.

Main Methods:

• Comparative evaluation of AutoDock Vina, HADDOCK, HDOCK, and RLDOCK.
• Assessment criteria included cognate docking success rate, blind docking performance, scoring accuracy, and ranking capabilities.
• Molecular dynamics simulations were employed to refine top-scoring docked poses.

Main Results:

• RLDOCK achieved a 70% success rate for top-scoring poses in cognate docking.
• All four methods failed to exceed a 50% overall success rate after pose refinement with molecular dynamics.
• All tested docking methods demonstrated poor performance in scoring potential evaluation, with AutoDock Vina showing mixed results (AUC 0.70) but poor performance in other metrics.

Conclusions:

• Current docking methods require significant optimization for accurate RNA-ligand interaction prediction.
• There is a clear need for developing and refining computational tools specifically designed for RNA-ligand docking.
• Improved RNA-ligand docking tools are essential for advancing the discovery of novel RNA-targeted cancer therapeutics.