Predicting RNA distance-based contact maps by integrated deep learning on physics-inferred secondary structure and...

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Recent advancements in protein structure prediction, like AlphaFold2 in CASP 14, highlight the potential for similar breakthroughs in RNA structure prediction.
• Accurate prediction of RNA secondary and tertiary structures is crucial for understanding biological function and disease mechanisms.
• Contact map prediction, a key component in protein structure prediction, has shown promise for advancing RNA structure modeling.

Purpose of the Study:

• To develop an accurate computational method for predicting RNA contact maps.
• To leverage deep learning, physics-based features, and evolutionary information for improved RNA structure prediction.
• To provide a tool that assists in the 3D structure prediction of structured RNAs.

Main Methods:

• Integration of deep learning algorithms with physics-inferred secondary structures.
• Utilization of co-evolutionary information derived from multiple sequence alignments.
• Application of multiple sequence alignment sampling to enhance prediction accuracy.
• Development of the SPOT-RNA-2D web server and standalone program.

Main Results:

• Achieved RNA contact-map prediction accuracy comparable to protein contact-map prediction.
• Demonstrated high accuracy for predicting long-range RNA contacts, particularly for RNAs with a high effective number of homologous sequences (Neff > 50).
• Validated the utility of predicted contact maps as distance restraints for 3D RNA structure prediction.

Conclusions:

• The developed method, SPOT-RNA-2D, significantly advances the field of RNA structure prediction.
• Accurate RNA contact map prediction is feasible and beneficial for 3D structure modeling.
• The tool is publicly available, facilitating further research and applications in RNA biology.