Machine Learning-Based Integration of Single-Cell and Bulk Transcriptome Reveals Coagulation Signature and Phenotypic Heterogeneity in Hepatocellular Carcinoma.

Primary liver cancer ranks as the third most lethal cancer globally, with hepatocellular carcinoma (HCC) being the most prevalent pathologic type. The liver plays a crucial role in maintaining normal coagulation function by synthesising, regulating and clearing coagulation factors and other bioactive substances involved in coagulation. Although several previous studies have proposed coagulation-associated prognostic models in HCC, the mechanisms at the single-cell level are not fully elucidated. In this study, the coagulation subtypes and their heterogeneity of HCC malignant cells were identified based on the coagulation-related genes collected from KEGG and GO databases. Through machine learning algorithms, we defined a coagulation gene signature at the single-cell level, based on which a coagulation-associated risk score (CARS) model was constructed in the TCGA-LIHC cohort. Integrating clinicopathological information and the CARS, a nomogram model was further developed for individualised prognostic assessment. Additionally, the mechanisms of prognostic differences among patients with divergent coagulation-associated risks were dissected through tumour signalling pathways, cellular communication and pseudotime trajectory analysis, while exploring the potential application of this risk assessment system in HCC treatment. In conclusion, the established CARS system accurately predicts prognosis, providing an important theoretical basis for precision treatment of HCC.