# Machine learning-based comprehensive analysis of m6A RNA methylation regulators in colorectal cancer: implications for prognosis, immune microenvironment, and immunotherapy response

## TL;DR

This study uses machine learning to analyze m6A RNA methylation regulators in colorectal cancer, showing they can predict patient survival and immunotherapy response.

## Contribution

A novel eight-gene m6A regulator signature was developed for CRC prognosis and immunotherapy prediction using machine learning.

## Key findings

- Random Forest achieved best performance with AUC of 0.895 in training and 0.847 in validation.
- Low-risk patients had higher CD8+ T cell infiltration and better predicted immunotherapy response.
- IGF2BP2 and METTL3 were top contributors to risk prediction based on SHAP analysis.

## Abstract

N6-methyladenosine (m6A) RNA methylation regulators have been implicated in colorectal cancer (CRC) progression. However, systematic evaluation using multiple machine learning approaches for prognostic prediction remains limited. This study aimed to develop and validate machine learning models for CRC prognosis based on m6A regulators and assess their potential for immunotherapy response prediction. We analyzed 1,047 CRC patients from TCGA and GEO databases (70% training, 30% validation). Twenty machine learning algorithms were systematically evaluated, with LASSO regression selecting optimal features from 27 m6A regulators. SHAP analysis provided model interpretability. Immune microenvironment characterization and immunotherapy response prediction were performed using established computational methods. LASSO regression selected eight m6A regulators (IGF2BP2, METTL3, HNRNPA2B1, METTL14, YTHDF2, VIRMA, FTO, ALKBH5) for model construction. Among 20 algorithms tested, Random Forest achieved optimal performance (training AUC = 0.895, validation AUC = 0.847). SHAP analysis identified IGF2BP2 (mean |SHAP| = 0.42) and METTL3 (mean |SHAP| = 0.36) as primary contributors to risk prediction. Risk stratification showed significant survival differences (HR = 2.41, 95% CI: 1.73–3.36, p < 0.001). Low-risk patients demonstrated enhanced immune infiltration with higher CD8+ T cells (17.8% vs. 10.2%, p < 0.001) and better predicted immunotherapy response rates (36.5% vs. 20.3%, p = 0.006). Our systematic machine learning analysis demonstrates that m6A regulators can effectively predict CRC prognosis and immunotherapy response. The eight-gene signature provides a practical tool for clinical risk assessment and treatment decision-making.