A comprehensive comparison of general RNA-RNA interaction prediction methods

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-RNA interactions are increasingly recognized as crucial functional elements in non-coding RNAs.
• Basepairing underlies the stability of these intermolecular RNA interactions, analogous to RNA secondary structures.
• Computational methods for predicting RNA secondary structure have been adapted for RNA-RNA interaction prediction.

Purpose of the Study:

• To conduct the first comprehensive comparison of 14 computational methods for predicting general intermolecular RNA basepairs.
• To evaluate method performance using experimentally validated fungal snoRNA-rRNA and bacterial sRNA-mRNA interaction datasets.
• To assess the impact of tool settings, sequence length, and multiple sequence alignment quality on prediction accuracy.

Main Methods:

• Compiled a dataset of 54 fungal snoRNA-rRNA and 102 bacterial sRNA-mRNA interactions.
• Tested and compared the performance of 14 distinct RNA-RNA interaction prediction tools.
• Analyzed the influence of various parameters, including sequence length and alignment quality, on prediction accuracy.

Main Results:

• Non-comparative, energy-based tools leveraging accessibility information achieved the highest accuracy for predicting short RNA-RNA interactions.
• Prediction accuracy significantly decreases across biologically diverse datasets and with increasing input sequence lengths.
• The developed interaction dataset is made publicly available for future research and benchmarking.

Conclusions:

• Current computational methods show limitations in maintaining high accuracy for RNA-RNA interaction prediction across varied biological contexts and longer sequences.
• The findings have implications for the feasibility of de novo transcriptome-wide RNA-RNA interaction searches.
• Further development is needed to improve the robustness and applicability of RNA-RNA interaction prediction tools.