INDEGRA

Overview

INDEGRA (INtegrated DEGRadation Analysis) corrects for RNA degradation artefacts in Oxford Nanopore direct RNA sequencing, enabling accurate transcript quantification and differential stability testing even from partially degraded samples. It estimates degradation directly from the read-end distribution, so no RIN score, spike-in, or paired reference sample is required.

Linux · macOS · Windows · CC-BY-NC-ND-4.0

Quick start

curl -LO https://github.com/Arnaroo/INDEGRA/releases/download/v1.2.0/INDEGRA-1.2.0-linux-x86_64-hpc.tar.gz
tar xzf INDEGRA-1.2.0-linux-x86_64-hpc.tar.gz
cd INDEGRA/
./INDEGRA --help

The problem it solves

Direct RNA sequencing on Oxford Nanopore platforms reads native molecules from the 3' poly(A) tail toward the 5' end. Intact molecules and 5'-truncated fragments from the same transcript therefore share a common 3' origin and differ only in where they start. Standard alignment-and-count pipelines cannot tell them apart, so a degraded sample produces systematic quantification bias and false differential expression that reaches up to 6 percent in controlled experiments and clusters into biologically interpretable but entirely artefactual functional categories.

A second bias affects any higher-eukaryote RNA-seq experiment: extensively overlapping isoforms produce reads that map to more than one transcript, and conventional tools allocate them proportionally to abundance, a rich-get-richer strategy that inflates shorter or already-abundant isoforms. INDEGRA models per-transcript degradation and uses the fitted profiles, rather than abundance, to guide ambiguous read assignment.

How it works

INDEGRA operates in stages on a standard transcriptome-aligned BAM file. It first adjusts annotated transcript boundaries to the empirical read-end distribution, then censors uninformative or artefactual reads. It estimates a per-transcript fragmentation rate by modelling random fragmentation as a Bernoulli process, which yields a truncated geometric read-length distribution with a simple closed-form maximum-likelihood estimator. Multi-mapped reads are then allocated to the transcript whose fragmentation profile they best fit, with a conservative fallback for genes that have no uniquely mapping support.

What it does not require

• No RIN score or degradation ladder; degradation is estimated from the read ends within each sample.
• No spike-in controls, though they can be used for validation.
• No paired undegraded reference sample.
• No change to the sequencing protocol or library preparation.
• No change to downstream workflows: outputs are standard BAM files and TSV tables compatible with DESeq2, edgeR, and limma-voom.