# Establishment of a m6A‐Related Molecular Pattern in the Prognosis and Immune Infiltration of Osteosarcoma Using Machine Learning and Experiments

## TL;DR

This study creates a new model using m6A-related genes to predict osteosarcoma prognosis and immune infiltration, validated through machine learning and experiments.

## Contribution

The study introduces a novel m6A-related gene model for osteosarcoma prognosis and immune infiltration prediction.

## Key findings

- A prognostic model using 14 m6A-related genes showed strong predictive power with AUC values up to 0.9091.
- The high-risk group had reduced immune infiltration and enriched malignancy-related pathways like E2F targets and MYC targets.
- A clinical nomogram was developed to support personalized treatment decisions in osteosarcoma.

## Abstract

To determine the prognosis of osteosarcoma, multiple predictive models have been constructed in recent years. Nevertheless, the model for N6‐methyladenosine (m6A)‐related genes, a critical subset of molecular regulators for osteosarcoma, has not been identified.

Gene expression matrices and clinical data were extracted from the GEO datasets GSE21257 and GSE16091. Randomly selected 70% of samples from GSE21257 were assigned as the training dataset, while the remaining 30% of samples from GSE21257 and all samples from GSE16091 were designated as the internal test and external test datasets, respectively. The predictive model was developed using elastic net–penalized Cox regression. Receiver operating characteristic (ROC) analysis, Kaplan–Meier analysis, and Wilcoxon′s tests were conducted in the training, internal test, and external test datasets to validate its efficacy. Additionally, a clinical nomogram was established for prognostic prediction. The expression of several signature genes was verified in osteosarcoma cell lines and clinical samples. In vitro experiments were performed to elucidate the impact of signature genes on the osteosarcoma phenotype. Immune infiltration analysis and gene set enrichment analysis (GSEA) were further integrated to validate the ability of the risk model to discriminate cancer characteristics.

A total of 110 m6A‐related and survival‐significant genes were identified from GSE21257. Among these, 14 genes were ultimately included in the prognostic model for osteosarcoma. ROC analysis showed that the AUC values in the training, internal test, and external test datasets were 0.8304, 0.9091, and 0.7123, respectively. Furthermore, the AUC values for predicting 1‐, 3‐, and 5‐year overall survival were 0.8827, 0.8709, and 0.7664, respectively, with an overall AUC of 0.8275. Under this framework, a clinical nomogram was successfully constructed. Notably, immune infiltration analysis revealed a reduced immune score in the high‐risk group. GSEA demonstrated enrichment of several well‐known malignancy‐related gene sets in the high‐risk group, including E2F target genes, MYC targets, mitotic spindle, and hypoxia‐related pathways, among others.

A prognostic model based on m6A‐related genes was developed, which exhibits strong efficacy in predicting the prognosis of osteosarcoma. Additionally, a robust clinical nomogram was generated, providing novel evidence to support clinical decision‐making and personalized treatment.