Long non-coding RNAs are routinely nominated as candidate disease regulators from steady-state RNA sequencing differential-expression catalogs, and a widely repeated explanation holds that short-lived lncRNAs are systematically missed because rapid decay depresses their steady-state abundance. A new analysis published in Molecular Genetics and Genomics asks how much of that proposed discovery bias RNA stability can actually account for, and concludes that the question cannot currently be answered with the half-life data on hand. The study, authored by Hidenori Tani and published on 26 October 2026 as article number 204 in volume 301 of the journal, frames the stability explanation not as wrong but as untestable at present, because the coverage of measured RNA half-lives is far too sparse to support a systematic evaluation.
The technical scope of the analysis is considerable. The author assembled 18 half-life strata drawn from 16 metabolic-labeling datasets spanning 13 human cell lines, and matched them against 17 uniformly processed disease cohorts from recount3, consisting of 16 TCGA tumor-versus-normal contrasts and one non-alcoholic fatty liver cohort, matching the half-life cell line to the tumor tissue where possible. Even before any stability question could be addressed, an abundance floor imposed by the pre-filter left only 24 percent of annotated lncRNAs testable in the design. Downsampling the breast cancer cohort showed how fragile differential-expression discovery is at small scale: at five samples per group, only 18 percent of the full-cohort differentially expressed lncRNAs were recovered, illustrating how heavily the candidate lists that feed the stability hypothesis depend on cohort size and detection thresholds. The upstream driver of this abundance floor is not resolved by the analysis.
At the heart of the stability explanation is an assumed coupling between rapid decay and low steady-state abundance. The analysis quantified that coupling directly. Across six tissue-matched half-life datasets covering four cell lines, the median half-life–abundance correlation was ρ = 0.195, with a cluster-bootstrap range of −0.003 to 0.293 that the author describes as descriptive rather than a formal interval given the small number of clusters; dropping the weakest tissue match lowered the median to 0.143. Those modest correlations indicate that half-life explains only a small share of abundance variation among measurable transcripts. Among the lncRNAs that do carry a measured half-life, the analysis found no excess exclusion of short-lived relative to long-lived transcripts from differential-expression catalogs. The author is careful, however, to flag the circularity embedded in that comparison: half-lives are measurable only where a transcript is expressed, and only in cell lines, so the comparison is confined to a subset that has already been selected on the very quantity at issue.
The decisive constraint is coverage rather than correlation strength. Only 10.0 percent of annotated lncRNAs have any measured half-life at all, and only 4.4 percent of the lncRNAs that survive the abundance filter do so. Counting tissue-matched cell lines only, the figure falls below 1 percent. In other words, the transcripts most relevant to disease-cohort analyses are precisely those least likely to have a decay measurement, and the cell lines in which half-lives have been profiled only rarely correspond to the tumor tissues in which the candidate lncRNAs were discovered. Under these conditions, any claim that stability systematically shapes which lncRNAs enter disease-regulator catalogs rests on extrapolation from a sliver of the annotation.
To make the gap concrete and reproducible, the study provides SIEVE, open-source code that produces these detectability figures for a given experimental design, allowing other researchers to compute how much of the lncRNA annotation is testable under their own cohort and half-life coverage. The paper also specifies the data the stability hypothesis would need to become testable: broader metabolic-labeling half-life measurements, particularly in cell lines matched to disease-relevant tissues, extending into the low-abundance range where short-lived transcripts are expected to sit. Until that coverage exists, the analysis concludes, the stability explanation for lncRNA discovery bias remains a proposed but currently untested account of how candidate lists are formed.
Subject of Research: Testability of the RNA stability explanation for long non-coding RNA discovery bias in disease transcriptomics
Article Title: The stability explanation for lncRNA discovery bias is not testable with current half-life coverage
Article References: Tani, H. (2026). The stability explanation for lncRNA discovery bias is not testable with current half-life coverage. Molecular Genetics and Genomics, 301(1), Article 204. https://doi.org/10.1007/s00438-026-02526-9
Image Credits: AI Generated
DOI: 10.1007/s00438-026-02526-9
Keywords: lncRNA, RNA half-life, RNA stability, differential expression, RNA-seq, TCGA, metabolic labeling, discovery bias, bioinformatics, transcriptomics, stability, explanation
