RNAmigos2: accelerated structure-based RNA virtual screening with deep graph learning.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Ribonucleic acid (RNA) represents a significant source of potential drug targets.
• Structure-based virtual screening (VS) is crucial for identifying drug candidates but faces challenges with RNA targets and large compound libraries.
• Current machine learning applications for RNA VS are limited by data scarcity and insufficient validation.

Purpose of the Study:

• To develop and validate a novel, data-driven VS pipeline specifically for RNA targets.
• To overcome the limitations of traditional molecular docking for large-scale RNA screening.
• To establish the efficacy of deep learning in structure-based RNA drug discovery.

Main Methods:

• Implementation of a VS pipeline using coarse-grained 3D modeling and RNA-specific self-supervision.
• Utilizing synthetic data augmentation to enhance model training.
• Employing a deep learning model for rapid screening of compound libraries against RNA structures.

Main Results:

• Achieved a 10,000-fold speedup compared to traditional docking methods.
• Consistently ranked active compounds within the top 2.8% across diverse RNA structures.
• Demonstrated robustness to variations in binding sites.
• Successfully screened novel RNA riboswitches in an in-vitro microarray (20,000 compounds) with a mean enrichment factor of 2.93 at 1%.

Conclusions:

• The developed deep learning pipeline offers a highly efficient and accurate method for structure-based RNA VS.
• This work represents the first experimental validation of structure-based deep learning for RNA VS.
• The approach significantly advances the potential for discovering drugs targeting RNA molecules.