Unraveling prognostic biomarkers in oral squamous cell carcinoma: An approach based on explainable artificial intelligence

Abstract

Oral cancer is among the top malignancies and the leading cause of death worldwide. Poor outcomes are attributed to local recurrence and distant metastasis of disease. There is an urgent need to identify the potential biomarkers that may help in prognostication and management of oral cancer. This study aimed to find potential prognostic biomarkers for oral squamous cell carcinoma (OSCC) using eXplainable artificial intelligence (XAI). After the curation of microarray data from GSE31056 (38 relapsed and 58 non-relapsed OSCC samples/ normal oral tissue samples), the application of XAI on Extreme Gradient Boosting algorithm machine learning (ML) models trained on binary classification datasets was employed. After successfully incorporating SHapley Additive exPlanations values into the ML models, 20 top significant genes associated with the relapse of OSCC were identified. The key genes included FAM49B, TTC39A, IFI16, ANGPTL4, HSPH1, GRIA2, SERF2 and others which contribute crucially to cell growth, cell invasion, apoptosis, disease progression, overall survival and disease-free survival. Further, a network of genes and their targeting microRNAs (miRNAs) revealed that miRNAs hsa-let-7b-5p, hsa-miR-27a-3p and hsa-miR-124–3p, had the highest interactions with genes. The predicted genes and miRNAs might be worthy prognostic markers and open the possibilities to understand the underlying pathways and recognize therapeutic targets for aggressive OSCC.