Prioritizing bona fide bacterial small RNAs with machine learning classifiers.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Bacterial small RNAs (sRNAs) regulate crucial cellular processes.
• RNA sequencing has identified numerous putative sRNAs across bacterial species.
• Experimental validation of all identified sRNAs is experimentally challenging.

Purpose of the Study:

• To develop and evaluate machine learning models for discriminating bona fide bacterial sRNAs from random sequences.
• To identify conserved genomic features predictive of bacterial sRNAs across species.

Main Methods:

• Applied five machine learning approaches to build twenty predictive models.
• Utilized seven genomic features, including secondary structure, promoter/terminator proximity, and distance to ORFs.
• Developed the sRNA Characterization Pipeline (sRNACharP) for automated feature calculation.

Main Results:

• The best model achieved high precision (100% at 10% recall) and outperformed previous methods on benchmark datasets.
• All seven features positively contributed to model performance.
• Identified conserved intrinsic genomic features of sRNAs across bacterial species.

Conclusions:

• Machine learning models can effectively prioritize bacterial sRNAs based on conserved genomic context features.
• This computational approach significantly reduces the need for extensive experimental validation.
• Developed models demonstrate species-independent predictive power for identifying bona fide sRNAs.