Integrated machine learning algorithms reveal a bone metastasis-related signature of circulating tumor cells in prostate cancer

Abstract

Abstract Bone metastasis is an essential factor affecting the prognosis of prostate cancer (PCa), and circulating tumor cells (CTCs) are closely related to distant tumor metastasis. Here, the protein-protein interaction (PPI) networks and Cytoscape application were used to identify diagnostic markers for metastatic events in PCa. We screened ten hub genes, eight of which had area under the ROC curve (AUC) values > 0.85. Subsequently, we aim to develop a bone metastasis-related model relying on differentially expressed genes in CTCs for accurate risk stratification. We developed an integrative program based on machine learning algorithm combinations to construct reliable bone metastasis-related genes prognostic index (BMGPI). On the basis of BMGPI, we carefully evaluated the prognostic outcomes, functional status, tumor immune microenvironment, somatic mutation, copy number variation (CNV), response to immunotherapy and drug sensitivity in different subgroups. BMGPI was an independent risk factor for disease-free survival in PCa. The high risk group demonstrated poor survival as well as higher immune scores, higher tumor mutation burden (TMB), more frequent co-occurrence mutation, and worse efficacy of immunotherapy. This study highlights a new prognostic signature, the BMGPI. BMGPI is an independent predictor of prognosis in PCa patients and is closely associated with the immune microenvironment and the efficacy of immunotherapy.