Development of a cancer metastasis-associated risk model via multi-machine-learning algorithms for prognostic risk evaluation and clinical application in oral squamous cell carcinoma

TL;DR

This study creates a 13-gene model to predict survival and identify treatment options for oral cancer patients based on metastasis-related genes.

Contribution

The first machine-learning-based prognostic model integrating EMT, anoikis, and basement membrane remodeling genes for oral cancer.

Findings

A 13-gene model effectively stratifies patients into high- and low-risk groups with strong survival prediction.

High-risk patients show altered immune infiltration, higher tumor purity, and lower immunotherapy response.

Seven potential therapeutic candidates were identified through computational drug screening.

Abstract

Oral squamous cell carcinoma (OSCC) represents a highly malignant form of cancer characterized by molecular heterogeneity and unsatisfactory treatment outcomes, with approximately 50% of patients experiencing local recurrence and distant metastasis following therapy. Given that metastasis is the most critical determinant of OSCC prognosis, enhancing the precision of clinical interventions and identifying therapeutic targets are of paramount importance. In view of this, this study is the first to develop a machine-learning-based prognostic model integrating epithelial-mesenchymal transition (EMT), anoikis, and basement membrane remodeling genes.

We systematically evaluated 78 algorithm and parameter combinations to identify a robust prognostic model, stratifying patients into High- and Low-risk groups. Kaplan-Meier survival curves and receiver operating characteristic (ROC) analyses were employed to evaluate the predictive performance of this model. Functional enrichment of differentially expressed genes (DEGs) between risk groups revealed key OSCC progression mechanisms. We further analyzed tumor mutation burden, immune microenvironment features, and identified candidate drugs through sensitivity prediction and molecular docking.

The identified 13-gene prognostic model effectively stratified patients into high- and low-risk groups, demonstrating strong predictive power for overall survival: the high-risk group exhibited worse prognosis. Mutation landscape demonstrated significant genetic variability within these model genes, which provided insights into the association between elevated tumor mutational burden and adverse prognostic outcomes. Immune landscape revealed a distinct tumor microenvironment: high-risk group exhibited altered immune cell infiltration, along with increased tumor purity, reduced ESTIMATE score and poorer anticipated response to immunotherapy. Finally, seven promising therapeutic candidates were identified through integrated computational drug screening.

We developed and validated a 13-gene prognostic model that integrates metastasis-related processes, improves survival prediction, and identifies therapeutic opportunities in OSCC.

The online version contains supplementary material available at 10.1186/s12967-025-07336-y.

Figures