Detecting riboSNitches with RNA folding algorithms: a genome-wide benchmark

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• RNA secondary structure prediction is complex, especially for messenger and non-coding RNAs.
• Detecting structural changes due to single nucleotide variants (SNVs), termed riboSNitches, is vital for understanding RNA function and phenotype.
• A large dataset of human RNAs and their riboSNitch status from a Parallel Analysis of RNA Structure (PARS) study offers a benchmark for evaluating prediction algorithms.

Purpose of the Study:

• To evaluate the performance of 11 RNA folding algorithms in predicting riboSNitches using the PARS dataset.
• To compare the efficacy of algorithms specifically designed for SNV impact prediction against general RNA structure prediction tools.
• To assess the impact of different prediction methodologies (e.g., base pairing probabilities vs. minimum free energy) on accuracy.

Main Methods:

• Benchmarking 11 RNA folding algorithms against a human genome-wide PARS dataset.
• Analyzing algorithm performance specifically on validated subsets of riboSNitches.
• Comparing predictions based on base pairing probabilities versus minimum free energy calculations.

Main Results:

• Algorithms specifically designed for SNV impact, including remuRNA, RNAsnp, and SNPfold, demonstrated superior performance on rigorously validated data.
• General RNA structure prediction algorithms (RNAfold, RNAstructure) showed improved accuracy when utilizing base pairing probabilities over minimum free energy.
• Overall aggregate performance across all algorithms for the full riboSNitch dataset was relatively low.

Conclusions:

• Specialized algorithms offer the most promising approach for accurate riboSNitch prediction.
• Base pairing probabilities are a more effective metric than minimum free energy for general RNA structure prediction algorithms in this context.
• Focusing on high-confidence predictions can significantly improve the evaluation of RNA structure prediction algorithms for riboSNitches.