Summary
A bioRxiv preprint analysing a PLP1-mutant Jimpy brain dataset finds that normalization choices and shared controls can affect which genes appear to move toward healthy levels after a rescue. The authors say such transcriptomic patterns can indicate broad state restoration without identifying which downstream gene mediates it.
A bioRxiv preprint posted on October 2, 2026, reports that normalization choices and shared controls can strongly affect which genes appear to move toward healthy levels in rescue transcriptomics. Analysing a PLP1-mutant Jimpy brain dataset (GSE277705), Hasan M Jamil and Alexander Gow argue that this pattern can support evidence of broad state restoration without showing which downstream transcript caused it.
In perturbation transcriptomics, researchers compare gene activity across conditions—for example, disease-model samples and samples in which a biological change has been introduced. A gene that shifts toward the control level after a rescue may be considered a candidate mediator. The preprint examines how that ranking can be affected by the way sequencing counts are normalized and by statistical coupling to shared control samples.
Normalization changes the apparent scale of rescue
The authors applied different normalization schemes to the same dataset. Counts per million (CPM) scales gene counts relative to the total sequencing reads in each sample; median-of-ratios (MOR) is another method for adjusting for differences in sequencing depth and composition. Such adjustments help compare samples, but the authors report that the resulting rescue classifications differed substantially.
Under the original negative-interaction criterion, between 47.6% and 76.0% of all expressed genes—and 77.6% to 86.6% of genes classed as disease-eligible—met the criterion across normalization schemes. With total CPM, the authors attribute much of this high fraction to a median wild-type treatment response of +1.016 standard deviations. Under MOR, that median response fell to +0.154 standard deviations, while the median response in Jimpy samples remained near zero.
Under MOR, genes that were increased in disease and genes that were decreased in disease showed similar enrichment for the rescue pattern: 77.6% and 74.0%, respectively. The authors interpret this mirror pattern as evidence that selection affects the ranking, but say it cannot by itself distinguish statistical coupling from genuine restoration of the underlying state.
Candidate-gene estimates were sensitive to sample choice
A slope decomposition estimated that mechanical coupling accounted for 48.7% of the observed effect. But removing just one of four untreated-Jimpy samples moved that estimate from 37.8% to 90.1%. Candidate-specific point estimates were 60.8% near TRIB3 and 38.6% near DDIT4. The estimated TRIB3-minus-DDIT4 difference remained positive across all four deletions, but its size varied from 8.6 to 27.7 percentage points.
The preprint also reports that oligodendrocyte-specific Perk deletion suppressed TRIB3 (R^JP = −0.817, 95% CI [−1.048, −0.587]) and DDIT4 (−0.292, [−0.419, −0.169]). In contrast, the authors report that 2BAct resolved neither transcript-specific rescue. Together, the analyses highlight a distinction important to interpreting perturbation studies: a rescue can be associated with broad movement toward a control-like state, while identifying a particular downstream transcript as the mediator requires evidence beyond its position in a rescue ranking.