Mitochondrial RNA modification in colorectal cancer: From single-cell analysis to machine learning-based risk modeling.

Abstract

The clinical relevance of mitochondrial RNA modification (MRM) in colorectal cancer (CRC), particularly its value for prognostic stratification, has not been fully defined. We integrated bulk transcriptomic profiles, machine learning-based model screening, single-cell analysis, and experimental validation to identify MRM score-associated prognostic genes and construct a CRC risk model. Seven CRC-related prognostic genes were selected: SPARCL1, MGP, PRELP, PALMD, TNS1, ARHGEF25, and PTGIS. These genes were incorporated into a risk signature with favorable prognostic performance, as supported by nomogram-based assessment. Gene Set Enrichment Analysis indicated that cytokine-related processes may participate in CRC progression. TNS1 showed the strongest positive association with natural killer cells (cor = 0.784, P < 0.05) and the strongest inverse association with type 17 T helper cells (cor = -0.279, P < 0.05). Database-based screening predicted 113 candidate compounds targeting CRC. Single-cell analysis further highlighted smooth muscle cells, epithelial cells, endothelial cells, T cells, and B cells as major cellular populations of interest. SPARCL1, MGP, PRELP, and PALMD increased during both early and late B cell maturation, whereas TNS1 and ARHGEF25 were highly expressed across broader cellular contexts. In conclusion, SPARCL1, MGP, PRELP, PALMD, TNS1, ARHGEF25, and PTGIS may represent MRM score-associated prognostic markers in CRC and require further study.