IRIS: A Machine Learning-Based Pose Reranking Tool for RNA-Ligand Docking

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• RNA molecules are crucial in cellular processes and disease, making them key therapeutic targets.
• Accurate prediction of RNA-ligand complex 3D structures via computational docking is vital for rational drug design.
• RNA-ligand docking is challenging due to RNA's flexibility and charged backbone, and existing tools like rDock struggle with accurate pose ranking.

Purpose of the Study:

• To develop and validate the Intelligent RNA Interaction Scorer (IRIS), a novel machine learning model to enhance RNA-ligand docking pose ranking.
• To improve the accuracy of identifying near-native ligand poses for RNA targets in drug discovery pipelines.

Main Methods:

• Developed IRIS, a regression model using physicochemical and interaction-based features.
• Trained IRIS on the largest dataset of experimental nucleic acid-ligand complexes available for an ML-based RNA docking tool (608 structures).
• Integrated IRIS with the rDock program to rerank ligand poses generated by rDock.

Main Results:

• IRIS significantly improved rDock's RNA-ligand pose ranking accuracy compared to using rDock scores alone.
• When using the best rDock protocol, IRIS increased the success rate of ranking a near-native pose within the top five from 64.6% to 78.0%.
• IRIS improved the top-ranked pose accuracy, with the correct pose ranked first in 59.8% of cases, up from rDock's 42.7%.

Conclusions:

• IRIS effectively enhances the accuracy of RNA-ligand docking pose ranking.
• The model can be seamlessly integrated into existing docking workflows to improve RNA-targeted drug discovery.
• IRIS offers a significant advancement for structure-based inhibitor design against RNA targets.