Constructing a mitochondrial-related genes model based on machine learning for predicting the prognosis and therapeutic effect in colorectal cancer

Abstract

The role of mitochondria in tumorigenesis and progression is has been increasingly demonstrated by numerous studies, but its prognostic value in colorectal cancer (CRC) remains unclear. To address this, we developed a mitochondrial-related gene prognostic model using 101 combinations of 10 machine learning algorithms. Patients in the high-risk group exhibited significantly shorter overall survival time. The high-risk group exhibited elevated tumor immune dysfunction and exclusion score, indicating diminished immunotherapy efficacy. To address the suboptimal treatment outcomes in these patients, we identified PYR-41 and pentostatin as potential therapeutic agents, which are anticipated to enhance therapeutic efficacy in the high-risk group. Additionally, four biomarkers (HSPA1A, CHDH, TRAP1, CDC25C) were validated by quantitative real-time PCR, with significant expression differences between normal intestinal epithelial cells and colon cancer cells. Our prognostic model provides accurate CRC outcome prediction and guides personalized therapeutic strategies.

Keywords: Biomarkers; Colorectal cancer; Machine learning; Mitochondrion; Prognosis.

Fig. 1

The flowchart graph of this study

Fig. 2

Identification of mitochondrial-related genes associated with the prognosis of CRC. A Volcano plot of DEMRGs. B Bar plot of GO enrichment analyses. C Bubble plot of KEGG enrichment analyses. D Forest plots of univariate Cox regression analysis. E Venn diagram of intersection genes of Apoptosis pathway genes and prognosis related MRGs. F Venn diagram of intersection genes of Chemical carcinogenesis–reactive oxygen species pathway genes and prognosis related MRGs. G Venn diagram of intersection genes of Fatty acid metabolism pathway genes and prognosis related MRGs

Fig. 3

The prognosis model of mitochondrial-related genes has a strong performance to predict the prognosis of CRC patients. A Ten machine learning algorithms, including 101 combinations, were used to construct prognostic models and calculate the C-index for each cohort in each model; B–F K–M survival curves for survival analysis between high- and low-risk groups in each cohort; G–K ROC curves were used to predict 1-year, 3-year and 5-year survival time

Fig. 4

Functional enrichment analysis of the differentially expressed genes between the high-risk and low-risk groups. A Volcano plot of the differentially expressed genes. B Bar plot of GO enrichment analyses. C Bubble plot of KEGG enrichment analyses. D, E GSEA enrichment analyses

Fig. 5

Comparison of differences in clinical factors and construction of nomogram. A Comparison the difference of clinical factors between high-risk and low-risk groups. B Univariate Cox regression analysis. C Multivariate Cox regression analysis. D–I K-M survival curves for survival analysis of independent risk factors between high-risk and low-risk groups. J–L Used the independent risk factors to predict 1-year, 3-year and 5-year survival time

Fig. 6

The risk score of the prognostic model for mitochondrial-related genes is associated with the tumor microenvironment in CRC. A Multiple methods were used to assess the distribution of immune cells between the two groups. B Correlation analysis between risk value and CD8 ⁺ T cell characteristic gene expression. C, D Analysis of differences in stromal score between the two groups and the correlation analysis between stromal score and risk score. E, F Analysis of differences in Tumor purity between the two groups and the correlation analysis between Tumor purity and risk score. G, H Analysis of differences in immune score between the two groups and the correlation analysis between immune score and risk score. (*** represents p < 0.001, ** for p < 0.01, * for p < 0.05, ns for no significance)

Fig. 7

The risk score of the prognostic model for mitochondrial-related genes can predict the therapeutic sensitivity in CRC. A, B The top 30 genes with mutation frequency in the high and low risk populations. C Comparison the TIDE score between the high and low risk groups. D Correlation analysis between TIDE score and risk score. E, F Comparison the difference of overall survival time between the high and low risk groups. G The CTRP database and the PRISM database were used for drug sensitivity analysis. (*** represents p < 0.001, ** for p < 0.01, * for p < 0.05, ns for no significance)

Fig. 8

Biomarkers screening and validation. A–D GEPIA database was used to analyze the expression of target genes. E–H TCGA database was used to analyze the expression of target genes. I–L GEPIA database was used to analyze the effect of target genes on the OS time of CRC patients. M–P TCGA database was used to analyze the effect of target genes on the OS time of CRC patients. Q–T qRT-PCR was used to verify the expression levels of target genes in human intestinal epithelial cells and human colon cancer cells. (**** represents p < 0.0001, *** represents p < 0.001, ** for p < 0.01, * for p < 0.05)