Hepatocellular carcinoma (HCC) is a highly heterogeneous cancer with a poor prognosis. During the development of cancer cells, mitochondria influence various cell death patterns by regulating metabolic pathways such as oxidative phosphorylation. However, the relationship between mitochondrial function and cell death patterns in HCC remains unclear. In this study, we used a comprehensive machine learning framework to construct a mitochondrial functional activity-associated programmed cell death index (MPCDI) based on scRNA-seq and RNA-seq data from TCGA, GEO, and ICGC datasets. The index signature was used to classify HCC patients, and studied the multi-omics features, immune microenvironment, and drug sensitivity of the subtypes. Finally, we constructed the MPCDI signature consisting of four genes (S100A9, FYN, LGALS3, and HMOX1), which was one of the independent risk factors for the prognosis of HCC patients. The HCC patients were divided into high- and low-MPCDI groups, and the immune status was different between the two groups. Patients with high MPCDI had higher TIDE scores and poorer responses to immunotherapy, suggesting that high-MPCDI patients might not be suitable for immunotherapy. By analyzing the drug sensitivity data of CTRP, GDSC, and PRISM databases, it was found that staurosporine has potential therapeutic significance for patients with high MPCDI. In summary, based on the characteristics of mitochondria function and PCD patterns, we used single-cell and transcriptome data to identify four genes and construct the MPCDI signature, which provided new perspectives and directions for the clinical diagnosis and personalized treatment of HCC patients.