Energy-Filtered Branching Alternatives Improve RNA Secondary Structure Recall

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Accurate RNA secondary structure prediction is crucial but challenging.
• Minimum Free Energy (MFE) methods have limitations in predicting all known helices.
• Multibranch loop thermodynamics significantly impact prediction accuracy.

Purpose of the Study:

• To investigate the impact of multiloop initiation and branching penalties on RNA structure prediction recall.
• To explore alternative RNA secondary structures beyond the standard MFE prediction.
• To develop an energy-based filtering method for identifying promising alternative structures.

Main Methods:

• Generated distinct optimal RNA structures by partitioning the branching-parameter space.
• Evaluated prediction recall for alternative structures using the Archive II dataset.
• Introduced an energy-based filtering method to select candidate structures based on adjusted residual energy.

Main Results:

• Many RNA sequences possess alternative structures with substantially higher recall than MFE predictions.
• The proposed energy-filtering method yields a compact set of candidate structures.
• The resulting ensemble offers a better recall-precision balance than Boltzmann sampling for helix classes.

Conclusions:

• Multiloop reparameterization offers significant predictive potential beyond standard MFE.
• Examining branching configurations provides structural insights missed by MFE.
• The energy-filtering approach provides a practical set of alternative structural hypotheses for further analysis.