For prognosis and therapeutic response, colorectal cancer (CRC) patients are highly variable across stages, correlated to high inter-tumor heterogeneity at the molecular level.1 Therefore, molecular subtyping needs to be determined for stratifying patients into distinct prognostic subgroups according to tumor biology.2 Metabolic reprogramming is an important cancer hallmark and confers a few cancer phenotypes.3 Metabolic landscape, comprising metabolites, proteins, and their interactions remains high variability across pan-cancer or even the same cancer type with distinct conditions.4,5 Thus, comprehending the transcriptional changes of metabolite-interacting proteins (MIPs) may provide valuable insights into promising therapeutic targets. In the present study, CRC patients were categorized into two subtypes, namely C1 and C2 based on prognostic MIPs, and each subtype owned a distinct clinical outcome. Meanwhile, an MIP-relevant risk score was generated and could accurately predict survival outcomes and personalized therapy for CRC based on bioinformatics mining. Additionally, a MIP-relevant gene, insulin-like growth factor-binding protein 3 (IGFBP3) was overexpressed in CRC tissues, and deficiency of IGFBP3 could facilitate the accumulation of intracellular reactive oxygen species and inhibit mitophagy in CRC cells. In summary, the current study proposed robust MIP-based molecular subtyping and relevant risk scores for guiding therapy selection and prognosis prediction of CRC patients, which might facilitate comprehension and clinical application for CRC metabolism heterogeneity.