Interpretable Multiomics Machine Learning Identifies GSDMB-Associated Epigenetic Repression and Reduced Immune Activity in Metastatic Colorectal Cancer

Authors

Costa, M. d. S., Marcelo, H. I., Kafouri, G. A., De Camargo, V.

Abstract

Background: Colorectal cancer (CRC) is a major cause of cancer-related mortality, with distant metastasis strongly associated with poor clinical outcomes. Integrating transcriptomic and epigenomic data through machine learning may improve the molecular characterization of metastatic CRC. Methods: We analyzed 518 primary tumors from the TCGA-COAD/READ cohort (436 non-metastatic [M0] and 82 metastatic [M1]) with matched RNA-seq and DNA methylation data. Five machine learning classifiers were evaluated for discrimination of metastatic status using stratified nested cross-validation. Model explainability was assessed using model coefficients for linear classifiers and SHAP-based feature importance for tree-based models. Differential expression, functional enrichment, methylation-expression correlation, immune-related transcriptional scoring, statistical mediation, and survival analyses were subsequently performed. Results: Integrated RNA-seq and DNA methylation showed the strongest discrimination between M0 and M1 tumors, with SVM reaching a ROC-AUC of 0.787 +/- 0.047. Six features - ARC, ASPDH, C13orf15, C4orf23, GPATCH3, and cg12040555 - ranked among the top 20 predictors across four models with consistent directions. cg10057218 methylation was inversely correlated with GSDMB expression (Spearman rho = -0.589) and increased in M1 tumors. Statistical mediation indicated a significant indirect association between M1 status and reduced GSDMB expression through cg10057218 methylation (indirect effect = -0.370; 95% CI [-0.519, -0.224]; proportion mediated = 82.1%). M1 tumors also exhibited reduced immune-related transcriptional activity. A 20-feature canonical molecular score separated M1 from M0 tumors (in-sample ROC-AUC = 0.918), while the corresponding Logistic Regression model achieved a nested cross-validated ROC-AUC of 0.781 +/- 0.046. Higher scores were associated with shorter overall survival (log-rank p = 1.67 x 10^-7). Conclusions: Transcriptomic and DNA methylation integration identified a consistently prioritized multiomics feature set associated with metastatic CRC. The findings highlight coordinated molecular and immune-related differences between M0 and M1 tumors and identify cg10057218-associated GSDMB repression as a candidate epigenetic feature of metastatic disease.

Preprint server: bioRxiv
The authors list and abstract were imported from bioRxiv on 06 Sep 2026.

Stats

1-terrible, 9-excellent. How would you rate this preprint? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.