Hepatocellular carcinoma (HCC) presents significant intertumoral heterogeneity, complicating prognosis and treatment. To address this, we performed an integrated single-cell RNA-sequencing analysis of HCC specimens using Seurat and identified malignant cells via Infercnv. Through a systematic evaluation of 101 machine learning algorithms used in combination, we developed tumor-cell-specific gene signatures (TCSGs) that demonstrated strong predictive performance, with area under the curve (AUC) values ranging from 0.72 to 0.74 in independent validation cohorts. Risk stratification based on these signatures revealed distinct therapeutic vulnerabilities: high-risk patients showed increased sensitivity to sorafenib, while low-risk patients exhibited enhanced responses to immunotherapy and transarterial chemoembolization (TACE). Pharmacogenomic analysis with Oncopredict identified four chemotherapeutic agents, including sapitinib and dinaciclib, with risk-dependent efficacy patterns. Furthermore, CRISPR/Cas9-dependency screening prioritized SRSF7 as essential for HCC cell survival, a finding confirmed by the identification of protein-level overexpression in tumors via immunohistochemistry. This multi-omics framework bridges single-cell characterization to clinical decision-making, offering a clinically actionable prognostic system that can be used to optimize therapeutic selection in HCC management.