CircRNA biogenesis machinery is activated but decoupled from circRNA output in hepatocellular carcinoma

Analysis archive for a single study of hepatocellular carcinoma (HCC) that asks two questions together: whether the circRNA biogenesis machinery behaves as a coordinated programme, and whether its activation determines circRNA output. Main findings. Fifty-seven of 60 circRNA biogenesis-related RNA-binding proteins (circRBPs) are upregulated in TCGA-LIHC, but RBP activation does not propagate to circRNA abundance: across the 15 circRBPs that also host circRNAs, transcript change averaged +0.261 log2 units while the change in their circRNAs averaged +0.033. Penalised Cox regression on primary tumours selects a single gene, SFPQ, which is prognostic (HR 1.55 per SD, P = 2.8e-6) but does not transport to two external cohorts (C-index 0.53 and 0.51). circRNA abundance is organised locus by locus: within-host-gene correlation is 0.296 in 17 paired tumours profiled for both circRNA and linear RNA, falling to 0.131 after adjustment for measured host-gene mRNA, a purity proxy and library burden. The prognostic signal in this axis is carried by host-gene mRNAs, chiefly NT5DC2, rather than by circRNA quantity. Contents. Analysis scripts (Python and R), 140 derived result tables underlying every reported number, an analysis registry that maps each sample count to its script and result file, vector figure sources with 300+ dpi renders, and the manuscript and supplementary materials. Provenance. This is a new, self-contained record for the present study. An earlier record (doi:10.5281/zenodo.23293107) mixes two projects that must not be confused: its versions 1.0–1.3.4 are an unrelated circRNA–miRNA–mRNA network analysis, and its versions 2.0.0–2.1.1 are earlier releases of this archive, deposited there as new versions of that older concept. The present record supersedes v2.1.1 and contains only this study, so its version list cannot send a reader to the other analysis. Not included. Public primary data are not redistributed; accessions are listed in README.md. One intermediate table (circRNA_gene_map.tsv) is deliberately withheld because its adjusted P-value column is misaligned with its feature rows; the authoritative version of the same statistics is fdr_all_features.tsv, which carries the Benjamini-Hochberg column.