A validation approach for RNA structures based on deep-learning 3D-structure predictions

Extending conkit-validate to RNA yields a mapless, contact-pattern validation method with performance similar to protein results, detecting RNA sequence register errors even at low-resolution density.

• Why it matters: RNA structure validation is difficult because experiments can be challenging to interpret and existing map-based tools require high-resolution data, leaving gaps when density is ambiguous.
• What they did: The approach compares contact patterns between experimental RNA models and structure predictions for the same RNA, using evolutionary covariance learned by state-of-the-art prediction models.
• The result: In a PDB scan, its detection rate does not depend on local density resolution, and case studies show it can identify register errors missed by map-based approaches.

AI summary of the abstract · may contain errors · not medical advice · read the paper.