← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-rna-quantification-featurecounts-counting / references / usage-guide.md

featureCounts - Usage Guide

Overview

featureCounts is a read counting program from the Subread package that assigns aligned reads to genomic features such as genes, exons, or promoters.

Prerequisites

# Conda (recommended)
conda install -c bioconda subread

# Ubuntu/Debian
sudo apt install subread

# From source
wget https://sourceforge.net/projects/subread/files/subread-2.0.6/subread-2.0.6-Linux-x86_64.tar.gz
tar xzf subread-2.0.6-Linux-x86_64.tar.gz
export PATH=$PATH:$(pwd)/subread-2.0.6-Linux-x86_64/bin

Quick Start

Tell your AI agent what you want to do: - "Count reads in my BAM files using featureCounts" - "Generate a gene count matrix from my aligned samples" - "Run featureCounts on stranded paired-end data"

Example Prompts

Basic Counting

"Run featureCounts on my aligned BAM files using the GRCh38 GTF annotation"

"Count reads at the gene level for all samples in my data/ directory"

Strand-Specific

"Count reads with reverse-stranded library prep (TruSeq Stranded)"

"Help me determine the strandedness of my RNA-seq data"

Advanced Options

"Count at the transcript level instead of gene level"

"Run featureCounts allowing multi-mapping reads"

What the Agent Will Do

  1. Identify the input BAM files and annotation GTF
  2. Determine library type (paired/single-end, stranded/unstranded)
  3. Run featureCounts with appropriate parameters
  4. Check the summary file for assignment rates
  5. Extract the count matrix for downstream analysis

Key Parameters

Parameter Description Default
-a Annotation file (GTF/GFF/SAF) Required
-o Output file Required
-t Feature type to count exon
-g Attribute for grouping gene_id
-s Strandedness (0/1/2) 0 (unstranded)
-p Paired-end mode Off
--countReadPairs Count fragments not reads Off
-T Number of threads 1
-M Count multi-mappers Off
-O Count overlapping features Off

Strandedness Guide

Library Type -s Value Examples
Unstranded 0 Older non-directional Illumina
Forward (read 1 sense) 1 Ligation/directional (e.g. NSR)
Reverse (read 1 antisense) 2 dUTP, Illumina TruSeq Stranded, NEBNext Directional (most common)

The wrong -s does not error; it silently assigns ~5-20% instead of ~80-90%. Determine it empirically (RSeQC infer_experiment.py, the STAR ReadsPerGene.out.tab columns 3 vs 4, or trying -s 0/1/2 and picking the highest Assigned), never from the kit name alone.

Typical Workflows

Standard RNA-seq (Illumina TruSeq Stranded)

featureCounts \
    -p --countReadPairs \
    -s 2 \
    -T 8 \
    -a Homo_sapiens.GRCh38.gtf \
    -o gene_counts.txt \
    sample1.bam sample2.bam sample3.bam

Single-End Unstranded

featureCounts \
    -s 0 \
    -T 8 \
    -a annotation.gtf \
    -o gene_counts.txt \
    *.bam

Do Not Use featureCounts for Transcript Quantification

featureCounts is a gene-level counter. Grouping by -g transcript_id with -O is NOT valid transcript quantification: isoform-ambiguous reads (most reads) are either discarded or double-counted across shared exons, so the per-transcript numbers are biased. For transcript-level abundance and DTE/DTU, use alignment-free EM quantification (rna-quantification/alignment-free-quant) and the transcript-aware testing in alternative-splicing/isoform-switching. Use -g overrides only for genuinely distinct meta-features (e.g. custom intervals), not isoforms.

Understanding Output

The main output file has these columns: 1. Geneid - Gene identifier from GTF 2. Chr - Chromosome(s) for the gene 3. Start - Start position(s) 4. End - End position(s) 5. Strand - Strand(s) 6. Length - Total exon length 7+ Sample columns - Raw counts per sample

Quality Metrics

Check the .summary file for: - Assigned - Reads successfully counted (target: >70%) - Unassigned_NoFeatures - Reads not overlapping any feature - Unassigned_Ambiguity - Reads overlapping multiple features - Unassigned_MultiMapping - Multi-mapped reads (if not using -M)

Tips