Metabolic enzymes play a central role in cancer metabolic reprogramming, and their dysregulation creates vulnerabilities that can be exploited for therapy. However, accurately measuring metabolic enzyme activity in a high-throughput manner remains challenging due to the complex, multi-layered regulatory mechanisms involved. Here, we present iMetAct, a framework that integrates metabolic-transcription networks with an information propagation strategy to infer enzyme activity from gene expression data. iMetAct outperforms expression-based methods in predicting metabolite conversion rates by accounting for the effects of post-translational modifications. With iMetAct, we identify clinically significant subtypes of hepatocellular carcinoma with distinct metabolic preferences driven by dysregulated enzymes and metabolic regulators acting at both the transcriptional and non-transcriptional levels. Moreover, applying iMetAct to single-cell RNA sequencing data allows for the exploration of cancer cell metabolism and its interplay with immune regulation in the tumor microenvironment. An accompanying online platform further facilitates tumor metabolic analysis, patient stratification, and immune microenvironment characterization.