Global or local? Predicting secondary structure and accessibility in mRNAs.

Area of Science:

• Molecular Biology
• Bioinformatics

Background:

• Determining messenger RNA (mRNA) structural properties is essential for understanding post-transcriptional gene regulation.
• Experimental data for mRNA structure is limited, necessitating accurate computational prediction methods.
• Existing RNA structure prediction tools vary in performance and parameterization, lacking standardized comparison metrics.

Purpose of the Study:

• To evaluate and compare different RNA structure prediction methods.
• To introduce a novel metric, structure accuracy, for assessing both global and local RNA structure predictions.
• To develop an optimized local folding approach for improved mRNA structure prediction.

Main Methods:

• Assessed prediction performance using transcriptome-wide enzymatic probing data and a curated set of cis-regulatory elements.
• Introduced 'structure accuracy' as a unified measure for evaluating prediction quality.
• Investigated the impact of locality parameters (maximum base pair span and window size) on prediction accuracy.
• Developed and validated the LocalFold method to address border effects in local structure prediction.

Main Results:

• Local folding prediction approaches demonstrated higher accuracy than traditional global methods.
• A maximum base pair span of 150 offered an optimal balance between predicting correct base pairs and minimizing long-range prediction errors.
• Adjusting window size to be larger than the maximum span effectively reduced errors at artificial sequence ends.
• The LocalFold method significantly improved robustness by mitigating border effects.

Conclusions:

• Local folding strategies provide more accurate mRNA structure predictions compared to global methods.
• Optimized parameter selection, specifically span and window size, is critical for enhancing prediction accuracy.
• LocalFold offers a robust and accurate solution for mRNA structure prediction, addressing limitations of existing approaches.