When Does Additional Information Improve Accuracy of RNA Secondary Structure Prediction?

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• RNA secondary structure is crucial for function, making accurate prediction a key challenge.
• Existing methods can be enhanced using auxiliary information, such as suboptimal RNA structures.
• Directability, derived from competing substructures, offers a new avenue for improving prediction accuracy.

Purpose of the Study:

• To introduce and evaluate novel features for RNA secondary structure prediction.
• To investigate the role of similarity (profiles) and topological features in improving prediction accuracy.
• To develop and test machine learning classifiers using these new features.

Main Methods:

• Introduced 'profiles' as a similarity measure for competing RNA substructures.
• Utilized topological data analysis (persistence landscapes) to derive topological features from profiles.
• Developed random forest classifiers incorporating these novel similarity and topological features.

Main Results:

• Similarity features are more impactful for RNA sequences with similar structures.
• Topological features are more beneficial for RNA sequences with dissimilar structures.
• Extensive testing demonstrated the sensitivity of classification accuracy and feature importance.

Conclusions:

• Novel similarity and topological features derived from competing RNA substructures significantly enhance secondary structure prediction.
• The utility of these features depends on the structural similarity of the training RNA sequences.
• This work provides new tools and insights for computational RNA structure analysis.