Exploring the gene-level ferroptosis–matrisome molecular axis of fibrosis in MASLD: a cross-cohort analysis separating biological signal from cell composition

This deposit supports the manuscript: "Exploring the gene-level ferroptosis–matrisome molecular axis of fibrosis in MASLD: a cross-cohort analysis separating biological signal from cell composition" (Banerjee S, Charoensupa R, Vanden Berghe W). The one-paragraph summary: Ten published fibrosis gene panels are statistically indistinguishable in two fully external cohorts; all converge on the matrisome, which accompanies fibrosis without carrying its measurement. Of nine ways to represent a liver biopsy's transcriptome, only a plain gene-level score (12,766 genes, no selection, frozen harmonisation) transfers across cohorts (10-seed ρ = 0.759 ± 0.111, F2+ AUROC 0.883 ± 0.068). One fifth of its signal is cell mix, measured per patient. With cell mix removed, the fibrosis score and the ferroptosis driver arm converge, meeting at FDXR (rank 17 of 12,766; slopes positive in 3/4 cohorts). The score's top genes are detectable in blood (CD163, TREM2); the signal is not inherited genetic risk and does not predict progression. Every claim is guarded: pre-registered gates, random-set nulls, a caught-and-disclosed leakage collapse, and a 131-row audit ledger. Contents: Complete analysis pipeline (R discovery stages, Python verification/refinement programme), PACKAGE_PROVENANCE.json file-level audit manifest, and reproducibility documentation. Private OSF project archive: https://osf.io/bwykj/.