Machine learning and molecular subtyping reveal the impact of diverse patterns of cell death on the prognosis and treatment of hepatocellular carcinoma.

Abstract

Programmed cell death (PCD) is a significant factor in the progression of hepatocellular carcinoma (HCC) and might serve as a crucial marker for predicting HCC prognosis and therapy response. However, the classification of HCC based on diverse PCD patterns requires further investigation. This study identified a novel molecular classification named PCD subtype (C1, C2, and C3) based on the genes associated with 19 PCD patterns, distinguished by clinical, biological functional pathways, mutations, immune characteristics, and drug sensitivity. Validated in 4 independent datasets, diverse cell death pathways were enriched in the C3 subtype, including apoptosis, pyroptosis, and autophagy, it also exhibited a highly infiltrative immunosuppressive microenvironment and demonstrated higher sensitivity to compounds such as Paclitaxel, Bortezomib, and YK-4-279, while C1 subtype was significantly enriched in cuproptosis and metabolism-related pathways, suggesting that it may be more suitable for cuproptosis-inducing agent therapy. Subsequently, utilizing the machine learning algorithms, we constructed a cell death-related index (CDRI) with 22 gene features and constructed prognostic nomograms with high predictive performance by combining CDRI with clinical features. Notably, we found that CDRI effectively predicted the response of HCC patients to therapeutic strategies, where patients with high CDRI were more suitable for sorafenib drug therapy and patients with low CDRI were more ideal for transarterial chemoembolization (TACE). In conclusion, the PCD subtype and CDRI demonstrate significant efficacy in predicting the prognosis and therapeutic outcomes for patients with HCC. These findings offer valuable insights for the development of precise, individualized treatment strategies.