Identification of Immuno-Diagnostic and Therapeutic Markers in Tumors and Inflammatory Diseases Based on Single-Cell and Machine Learning Methods.

Background: Chemotherapy is the mainstay of systemic treatment for colorectal cancer (CRC), but drug resistance and tumor heterogeneity markedly limit survival. Mitochondrial metabolism is increasingly recognized as a regulator of tumor progression, immune contexture, and therapeutic response. However, the prognostic relevance of mitochondrial metabolism-related transcriptional patterns and their relationships with chemotherapy-response signatures and immune features in CRC remain incompletely understood. Methods: Clinical and transcriptomic data from The Cancer Genome Atlas colorectal cancer cohort (TCGA-CRC) and Gene Expression Omnibus cohorts GSE17537 and GSE104645 were analyzed. A curated set of mitochondrial metabolism-related genes (MMRGs) were used to calculate an MMRG score, whereas genes upregulated in chemotherapy-sensitive cases in GSE104645 were used to derive a drug-sensitivity gene-set score by single-sample gene set enrichment analysis (ssGSEA). Weighted gene coexpression network analysis (WGCNA), differential expression analysis, univariate Cox regression, least absolute shrinkage and selection operator regression, and multivariate Cox regression were used to construct a five-gene RiskScore. Prognostic associations were assessed using Kaplan-Meier curves, time-dependent receiver operating characteristic curves, external validation, and subgroup analyses. Immune-cell enrichment and pRRophetic-predicted half-maximal inhibitory concentration values were evaluated, and CD36 was examined using in vitro phenotypic assays. Results: Tumors showed lower MMRG scores than normal tissues, and higher MMRG scores were associated with better overall survival (OS). The drug-sensitivity gene-set score was higher in tumors but was not significantly associated with OS. A five-gene signature (CD36, CPA3, FCRL2, POU2F2, and S100B) generated a RiskScore that separated OS outcomes in the TCGA training, testing, and overall cohorts, as well as GSE17537, although its time-dependent discrimination varied across cohorts and follow-up periods. Similar survival associations were observed within subgroups defined by stage, age, and sex; however, independent or incremental prognostic value was not evaluated. High-risk tumors exhibited altered immune-cell enrichment, and the RiskScore was correlated with pRRophetic-predicted drug IC50 values after Benjamini-Hochberg correction. CD36 expression was higher in SW1116 than in NCM460 cells, and its knockdown inhibited proliferation, migration, and invasion. Mitochondrial function was not directly measured in these experiments. Conclusion: The five-gene RiskScore was associated with OS, immune-cell enrichment, and in silico drug-sensitivity estimates in CRC. It may serve as a candidate tool for prognostic stratification, but its incremental clinical value and treatment-predictive utility require prospective validation. The CD36 knockdown experiments support a tumor-promoting phenotype in SW1116 cells but do not establish dependence on mitochondrial metabolic reprogramming.