Interpretable machine learning driven biomarker identification and validation for prostate cancer.

Background: Prostate cancer (PCa), a common malignancy among men globally, requires the identification of biomarkers for early diagnosis and predicting progression. This study aimed to identify the key genes involved in the occurrence and development of PCa.

Methods: Leveraging data from the Gene Expression Omnibus (GEO) database, this study integrated multi-chip datasets, conducting differential expression analysis and enrichment analysis to pinpoint PCa-related genes. Subsequently, machine learning models were constructed using least absolute shrinkage and selection operator (LASSO) regression, support vector machine (SVM), and random forest (RF) methods. The optimal model was selected for further study and the contribution of related genes was explained using SHapley Additive exPlanations (SHAP) analysis. Furthermore, gene set enrichment analysis (GSEA) and immune cell infiltration analysis were utilized to uncover the underlying molecular mechanisms.

Results: In this study, 222 differentially expressed genes (DEGs) were identified and found to be enriched in functions and pathways potentially associated with PCa. Using multiple machine learning models, eight PCa-related core genes (, , , , , , , and ) were identified. The most accurate RF model was selected for further study with SHAP analysis, which also revealed the contribution of the above genes. GSEA and immune cell infiltration analysis uncovered distinctions between PCa and normal tissues.

Conclusions: This study offered potential biomarkers and a theoretical basis for the diagnosis and treatment for PCa.