2.2. ScRNA-Seq and Bulk RNA-Seq Data Acquisition and Preprocessing

ScRNA-seq data with a reading depth of 10× genomics (including mRNA expression profiles) of 3200 cells from four HCC samples were obtained from the {"type":"entrez-geo","attrs":{"text":"GSE146115","term_id":"146115"}}GSE146115 dataset in the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) (accessed on 15 July 2021) database and were performed to identify differentiation-related genes. Then, the “Seurat” package of R (v3.5.243) was used to process the scRNA-seq data, and the Percentage Feature Set function was used to filter the cells with ≥5% of mitochondria-expressed genes. Next, cells with a gene number <100 and a sequencing number <50 were filtered using correlation analysis, which was conducted to elucidate the relationship between sequencing depth per cell and total intracellular RNA sequences.

After filtering low-quality cells, the scRNA-seq gene expression dates were normalized by the LogNormalize function, and 1500 feature genes with high cell-to-cell variation were identified using the Find Variable Features method. In addition, gene expression profile information and corresponding clinical information of RNA-seq data from 50 paracancerous tissues and 374 HCC samples were extracted from a publicly available genetic database of cancer patients, TCGA (https://tcga-date.nci.nih.gov/tcga) (accessed on 18 July 2021), as a training dataset. Simultaneously, 260 HCC samples were obtained from the ICGC (https://dcc.icgc.org/donor.LIRI-JP.tsv.gz) (accessed on 20 July 2021) dataset as a validation dataset, among which 28 patients did not have detailed clinical data (e.g., survival time = 0 or unknown; absence of pathological diagnosis) was removed from the study.