Processed cross-species liver single-cell RNA-seq atlas (468,161 cells) with scVI integration and lineage annotations

This resource contains the processed, integrated cross-species liver single-cell RNA-seq atlas accompanying the study "A candidate ATF3–myeloid framework linking MASLD fibrosis programs to spatial immunotherapy-associated states in HCC". The atlas integrates 468,161 liver cells (377,338 mouse; 90,823 human) from eight analysis cohorts spanning seven public GEO accessions, covering acute and chronic chemical injury (APAP, CCl4), diet-induced and alcoholic steatohepatitis models (WD, WDA, WD+CCl4, HFD+CCl4, CDHFD-NASH), NASH-HCC, and human cirrhosis. Cells were quality-controlled, reduced to 4,014 highly variable genes, and integrated with scVI (batch key = source cohort; 30-dimensional latent space). The object provides per-cell lineage annotations (10 liver lineages including Kupffer cells and infiltrating macrophage subsets, assigned by cluster-level voting of canonical marker-panel scores), leiden clustering, UMAP/PCA embeddings, and raw counts. Included files: the annotated AnnData object (h5ad), the trained scVI model, the atlas post-processing/annotation script, the KC-1302 program gene universe with expected directionality, and per-patient TCGA-LIHC signature scores merged with clinical covariates and survival endpoints. All source data are publicly available under GEO accessions GSE136103, GSE145086, GSE189606, GSE216836, GSE228305, GSE232182, GSE269059.