Detecting and characterizing microRNAs of diverse genomic origins via miRvial

Abstract

MicroRNAs form an essential class of post-transcriptional gene regulator of eukaryotic species, and play critical parts in development and disease and stress responses. MicroRNAs may originate from various genomic loci, have structural characteristics, and appear in canonical or modified forms, making them subtle to detect and analyze. We present miRvial, a robust computational method and companion software package that supports parameter adjustment and visual inspection of candidate microRNAs. Extensive results comparing miRvial and six existing microRNA finding methods on six model organisms, Mus musculus, Drosophila melanogaste, Arabidopsis thaliana, Oryza sativa, Physcomitrella patens and Chlamydomonas reinhardtii, demonstrated the utility and rigor of miRvial in detecting novel microRNAs and characterizing features of microRNAs. Experimental validation of several novel microRNAs in C. reinhardtii that were predicted by miRvial but missed by the other methods illustrated the superior performance of miRvial over the existing methods. miRvial is open source and available at https://github.com/SystemsBiologyOfJianghanUniversity/miRvial.

Figure 1.

Major steps and flow chart of miRvial. It shows the steps on how raw sequencing reads are initially processed, including 3′ adapter trimming (step 1); the remaining reads are aligned to a reference genome in step 2, where loci with sufficient reads are merged (red brackets in step 3) and extracted for secondary structure analysis (steps 4 and 5). miRvial identifies miRNA-like hairpins using a representation of three features based on secondary structures (steps 6 and 7). It searches for miRNA duplexes with the characteristic ∼2-nt 3′-overhangs using the alignment of sequencing reads (step 8). Application of the steps lead to a list of candidate miRNAs in the reference genome.

Figure 2.

Variation of miRNA hairpins measured in three structural features. The first column lists model species. The hairpin length and the size of maximum bulge are in the unit of the number of nucleotides. The stem length is in the unit of the number of base pairs. The central mark of a box plot represents the median value in an organism, while the edges of the box indicate the 25th and 75th percentiles. The plotted whiskers correspond to approximately 99% coverage of data points, and outliers are plotted as individual points outside the whiskers. Values in parenthesis represent the values for the three parameters set up in miRvial. miRNA information was retrieved from miRBase version 21.

Figure 3.

A schematic graphic output from miRvial. miRvial provides graphic output to assist visual inspection and selection of genuine miRNAs. Shown is an example of a miRNA detected by miRvial. (A) The upper panel gives the number of candidate miRNAs miRvial reported; one in this case shown. The panel shows the unique genomic locus of the miRNA, along with the predicted RNA secondary structure represented in parentheses, and the folding energy in a negative value. The lines below show the aligned sequencing reads, followed by the number of reads (# reads), the length of the unique read (length), and the number of mappable loci of the unique read (#. loci). (B) The lower-left plot shows the hairpin structure, where putative mature miRNAs are highlighted in color (blue from 5p-arm and red from 3p-arm). (C) The lower-right plot shows the distribution of reads in the predicted precursor sequence.

Figure 4.

Comparison of the methods for miRNA prediction. (A) The known miRNAs in mouse, fruitfly, moss, algae and A. thaliana, as annotated in miRBase or previous studies (see the main text), are used as true positives for comparison. (B) The performance of miRvial, miRDeep2, miRTRAP and MIReNA on three animal species. (C) The performance of miRvial, miRDeep-P, miRPlant and miRA on plant organisms. Sensitivity is true positives divided by the number of known miRNA. Precision is true positives divided by the number of predicted positive miRNAs. F1 score is 2 × precision × sensitivity divided by (precision + sensitivity), which is the harmonic mean of precision and sensitivity.

Figure 5.

RT-PCR validation of five novel miRNAs in Chlamydomonas reinhardtii. The appearances of miRNAs on the 3p- and 5p-arms of five novel miRNA candidates are tested by RT-PCR. The PCR products on 4% agarose gels are marked as below. 3P and 5P: 3p-arms and 5p-arms of miRNAs, respectively; std1: a synthetic exogenous reference gene as a positive control; M: 20 bp DNA marker; the arrow points to the location of 80 bp on the marker; the genomic loci of the miRNAs are listed below the figures.