Immune related gene signature derived from integrate machine learning algorithm predicts outcomes of HNSC patients

Abstract Background: Due to the highly heterogeneous of tumor, head and neck squamous cancer (HNSC) patients are in variable immunotherapeutic response and have different clinical outcomes. We since develop the immune related gene signature (IRGS) using a machine learning based integrative procedure for distinguishing the immune microenvironment subtype of diverse HNSC patients and then help improve the outcomes of HNSC. Methods: This study integrate 10 machine learning algorithms to 111 combination for screening out the best immune related gene signature (IRGS) based on 4 multicenter cohorts. Survival analysis, multivariate Cox regression analysis, and decision curve analysis (DCA) were employed to assess the performance of IRGS. Gene Ontology(GO) and Kyoto Encyclopedia of Genes and Genomesenrichment (KEGG) analyses were conducted to evaluate the potential biological functions and mechanisms of IRGS. Results: Through the integrated machine learning algorithms, we constructed a 17-IRG signature, which demonstrated to be an excellent prognostic model in all cohorts and displayed better efficiency when compared with other 68 published prognostic signatures. IRGS exhibits a strong negative correlation with immune characteristics. The IRGS low group demonstrates increased immune infiltration and heightened sensitivity to immunotherapy, whereas the IRGS high group exhibits a higher frequency of deletion mutations in tumor suppressor genes. Besides, considering IRGS high patients insensitive to immunotherapy and their poor prognosis, we scheduled an agents screening strategy and selected dasatinib as the most potential target drug for IRGS high patients. Conclusions: IRGS was demonstrated excellent prognostic efficiency and offer a more precise selection for assessing pre-immune efficacy, which will help improve clinical outcomes for individual HNSC patients.