miRDeep2 Analysis - Usage Guide

Overview

Discover novel miRNAs and quantify known miRNAs with miRDeep2, which scores genome-mapped read stacks against the Dicer/Drosha biogenesis signature (a sharp mature arm, a lower-abundance star arm with ~2-nt 3' overhangs, a stable hairpin fold, and a randfold p-value). The key judgment is that a miRDeep2 score is a structural and expression hypothesis, not a validated miRNA: novel discovery is intrinsically high false-positive, and the classic false positives are tRNA and rRNA fragments whose hairpins throw scoring read stacks. There is no universal score cutoff - survey.pl reports signal-to-noise and an estimated FDR across cutoffs, and the analyst picks and reports one. Reach for miRDeep2 only when de novo discovery is actually needed; otherwise quantify known miRNAs with a faster tool.

Prerequisites

conda install -c bioconda mirdeep2 bowtie viennarna
# miRDeep2 uses bowtie 1 (not bowtie2)

# miRBase references (pin the version)
wget https://www.mirbase.org/download/mature.fa
wget https://www.mirbase.org/download/hairpin.fa

Quick Start

Tell your AI agent: - "Discover novel miRNAs in my collapsed small RNA-seq reads with miRDeep2" - "Set up the human genome bowtie 1 index and run mapper.pl" - "Quantify known human miRNAs only, skipping discovery" - "Filter my novel candidates against tRNA and rRNA loci" - "Use survey.pl signal-to-noise to choose a score cutoff"

Example Prompts

Novel miRNA Discovery

"Run miRDeep2 discovery on my collapsed FASTA with human and mouse miRBase references"

"My novel candidates overlap tRNA loci - how do I reject those false positives?"

"Which score cutoff should I use, and how do I justify it from the survey output?"

Quantification

"Quantify known human miRNAs with quantifier.pl without running discovery"

"Why do quantifier.pl and miRDeep2.pl give different counts for the same miRNA?"

Results Analysis

"Parse the result CSV and keep candidates above my chosen signal-to-noise cutoff"

"Check star-arm read support and randfold p-value for my top novel calls"

What the Agent Will Do

Build the bowtie 1 genome index and confirm reads are adapter-trimmed and collapsed
Run mapper.pl to align collapsed reads and emit the ARF alignment file
Run miRDeep2.pl with same-species and related-species miRBase references for discovery and quantification (or quantifier.pl for known-only)
Choose a score cutoff from survey.pl signal-to-noise / FDR and report it
Filter novel candidates against tRNA/rRNA/snoRNA loci and require star-arm support and reproducibility before trusting them

Tips

A miRDeep2 score is a hypothesis, not a validated miRNA - novel calls need orthogonal validation (Northern/qPCR, Dicer/Drosha-knockdown loss of signal, conservation)
There is no universal cutoff; the "score > 10 = high confidence" rule is folklore - use survey.pl signal-to-noise (Friedländer used the lowest cutoff giving SNR >= 5) and report your choice
The classic false positive is a tRNA/rRNA fragment hairpin - always intersect novel candidates against structured-RNA annotations
miRDeep2 needs a reference genome and bowtie 1 (not bowtie2), and pre-collapsed reads with the _xN count
Pin the miRBase version; convert U->T and strip header whitespace in reference FASTAs
Provide a related-species mature FASTA for conservation evidence; use 'none' only if no relative is available
If you only need known-miRNA counts, run quantifier.pl (or mirge3) and skip the expensive discovery engine
Judge novel calls against the community criteria (Ambros 2003), where consistent 5' processing of both arms is the single most discriminating signal - deliver a criteria table, not just a score ranking
miRDeep2 is genome-anchored; for a non-model animal with no assembly (or single-cell), use a genome-free tool such as Mirnovo

smrna-preprocessing - Adapter trimming and read collapsing
mirge3-analysis - Faster known-miRNA and isomiR quantification
differential-mirna - Differential expression of the count matrix
trf-pirna-profiling - tRF/piRNA biology that otherwise appears as false positives
genome-annotation/ncrna-annotation - Annotate tRNA/rRNA loci to filter candidates