Prediction of novel precursor miRNAs using a context-sensitive hidden Markov model (CSHMM)

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Small non-coding RNAs, particularly microRNAs (miRNAs), are crucial for biological regulation.
• While 706 human miRNAs are known, many more are predicted to exist in the genome.
• Accurate identification of novel miRNAs is essential for understanding gene regulation.

Purpose of the Study:

• To propose and evaluate a context-sensitive Hidden Markov Model (CSHMM) for representing miRNA structures.
• To develop an ab initio method for miRNA identification using the CSHMM in conjunction with filters.
• To assess the performance of the CSHMM against existing miRNA prediction tools.

Main Methods:

• Developed a context-sensitive Hidden Markov Model (CSHMM) trained on known human miRNA sequences.
• Evaluated the CSHMM classifier using cross-validation, known miRNA datasets, and non-coding RNA datasets.
• Integrated the CSHMM into a pipeline with filters for EST matches and Drosha cutting sites for genome-wide scanning.

Main Results:

• The CSHMM demonstrated superior performance compared to miPred and CID-miRNA.
• The CSHMM pipeline successfully identified 49 novel miRNAs on human chromosome 19.
• The pipeline also discovered 308 novel miRNAs from deep sequencing of human leukocyte small RNA sequences.

Conclusions:

• The CSHMM is a valuable tool for miRNA discovery, applicable to individual sequences and genome-wide analysis.
• The developed pipeline, combining CSHMM with filters, shows significant promise for ab initio identification of novel miRNAs.
• This approach enhances the discovery of new microRNAs, contributing to a deeper understanding of gene regulation.