Knotted artifacts in predicted 3D RNA structures.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Topological knots are rare in experimentally determined RNA structures within the Protein Data Bank.
• Recent findings include the first trefoil knots and lasso conformations in experimental RNA, but these remain exceptional.
• Computational RNA structure prediction algorithms, especially machine learning-based ones, often produce knotted RNA folds and structural entanglements.

Purpose of the Study:

• To analyze topological knots and structural entanglements in RNA models submitted to the CASP15 competition.
• To identify the prediction methods associated with the generation of these knotted and entangled RNA conformations.
• To investigate the structural features that contribute to RNA entanglement susceptibility in predictive models.

Main Methods:

• Analysis of all 3D RNA structure prediction models from the CASP15 competition.
• Topological analysis to identify and classify knots and entanglements in predicted RNA structures.
• Correlation of identified topological features with the specific prediction algorithms used.

Main Results:

• CASP15 RNA models exhibit various types of topological knots and structural entanglements.
• Machine learning-based RNA structure predictors are more prone to generating these artifacts than traditional methods.
• Specific structural characteristics are associated with increased susceptibility to entanglement during prediction.

Conclusions:

• Predictive algorithms can introduce topological artifacts like knots and entanglements into RNA models.
• There is a need for improved evaluation metrics in RNA structure prediction to detect and avoid such algorithmic artifacts.
• Careful assessment of predicted RNA models is crucial to prevent the publication of algorithmically generated structures with unusual entanglements.