Towards an integrated pipeline for the in-silico prediction of plant microRNAs and their precursors

MicroRNAs (miRNAs) are endogenous, small (~ 20 nt), single-stranded, non-coding RNAs that result from the processing of transcribed precursor hairpin structures. They are increasingly recognized as playing crucial roles as post-transcriptional antisense regulators of gene expression through regulation of mRNA stability or translational efficiency. The detection of homologs of known miRNAs through comparative genomic approaches has proved relatively tractable. However, the ab-initio prediction of potentially lineage-specific miRNA precursors through computational methods poses several additional difficulties, not least the fact that not all thermodynamically plausible transcribed hairpins are processed to yield mature miRNAs. We have developed a Support Vector Machine that considers up to 78 features associated with the primary and secondary structures and thermodynamic characteristics of candidate hairpin structures. Our SVM is highly specific in the discrimination of true miRNA precursors from “spurious” hairpins with levels of false positive predictions that are low relative to comparable methods. We also show how our SVM functions as part of an in-silico pipeline for the prediction of novel miRNA precursors in plant genomes.