Development of an integrated machine learning-based approach utilizing NK cell marker genes for prostate cancer prognosis and treatment response characteristics

Abstract

Background Despite prostate cancer's (PCa) highly variable behavior and unclear response to immunotherapy, the importance of NK cells isn't comprehensively studied. Our study aimed to use a robust computational framework to consider NK cell marker gene signatures (NKCMGS) from 1,072 global PCa patients, intending to establish a reliable biomarker that can prognose and predict immunotherapy response. Methods NK cell-related biomarkers were studied in PRAD patients from six worldwide cohorts, creating a reliable NKCMGS biomarker using 101 genes and varied machine learning techniques. NKCMGS was further analyzed immunologically, providing new immunotherapy response and prognosis perspectives. Results The NKCMGS integrated 13 key genes, effectively classifying patients into high- and low-risk groups. Survival curves drawn from NKCMGS scores, age, T stage, and Gleason scores, established the reliable prognostic trait of NKCMGS. Biologically, high-scored NKCMGS indicated enhanced fatty acid and β-alanine metabolism pathways, while low scores showed enrichment in DNA repair and replication, homologous recombination, and cell cycle pathways. Moreover, low-risk patients demonstrated higher drug sensitivity, thus suggesting the potential of NKCMGS in predicting immune checkpoint inhibitor effectiveness. Conclusion Our robust machine learning framework, integrated with NKCMGS, show significant potential for providing personalized risk assessment and valuable treatment strategies for PCa patients.