Classification of RNA structure change by 'gazing' at experimental data

Area of Science:

• RNA structure and dynamics
• Computational biology
• Bioinformatics

Background:

• Mutations in RNA can alter structure and function, termed riboSNitches.
• Predicting riboSNitches requires comparing wild-type and mutant RNA structures.
• Selective 2' Hydroxyl Acylation by Primer Extension (SHAPE) data is key for validation.

Purpose of the Study:

• To develop a computational classifier that mimics expert analysis of SHAPE data for riboSNitch identification.
• To improve the accuracy of RNA structure prediction by incorporating SHAPE reactivity features.

Main Methods:

• Utilized expert consensus on SHAPE data to train a classifier.
• Employed dynamic time warping and seven other features to identify riboSNitches.
• Developed the classSNitch R package for predicting riboSNitches.

Main Results:

• Achieved strong quantitative agreement between human experts in identifying riboSNitches from SHAPE data.
• The classSNitch classifier demonstrated high accuracy (AUC > 0.8) in predicting riboSNitches.
• Incorporating wild-type SHAPE reactivity features improved thermodynamic structure predictions.

Conclusions:

• The classSNitch classifier accurately predicts riboSNitches by simulating expert analysis.
• Improved RNA structure prediction holds potential for precision medicine applications.
• The classSNitch R package is publicly available for use.