Measuring covariation in RNA alignments: physical realism improves information measures

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Non-coding RNA function is often structure-dependent.
• RNA secondary structure prediction relies on detecting co-evolutionary patterns in sequence alignments.
• Measuring positional covariation is key to structure prediction.

Purpose of the Study:

• To rank and evaluate various covariation measures for RNA secondary structure prediction.
• To identify which measures are most effective for improving prediction accuracy.

Main Methods:

• Comparison of multiple covariation measures, from simple mutual information to advanced methods.
• Assessment of measures based on their ability to predict RNA secondary structure.
• Evaluation of incorporating structural information like indels and stacking.

Main Results:

• Mutual information is commonly used but not always the most effective.
• Measures incorporating structural information, such as indels and stacking, improve prediction accuracy.
• The modified RNAalifold covariation measure, including stacking, performed best.

Conclusions:

• Physically realistic covariation measures lead to improved RNA secondary structure predictions.
• The findings can enhance current and future RNA structure prediction software.
• The best-performing measure integrates stacking information.