Neuroendocrine prostate cancer (NEPC) typically implies severe lethality and limited treatment options. The precise identification of NEPC cells holds paramount significance for both research and clinical applications, yet valid NEPC biomarker remains to be defined. Methods: Leveraging 11 published NE-related gene sets, 11 single-cell RNA-sequencing (scRNA-seq) cohorts, 15 bulk transcriptomic cohorts, and 13 experimental models of prostate cancer (PCa), we employed multiple advanced algorithms to construct and validate a robust NEPC risk prediction model. Results: Through the compilation of a comprehensive scRNA-seq reference atlas (comprising a total of 210,879 single cells, including 66 tumor samples) from 9 multicenter datasets of PCa, we observed inconsistent and inefficient performance among the 11 published NE gene sets. Therefore, we developed an integrative analysis pipeline, identifying 762 high-quality NE markers. Subsequently, we derived the NE cell-intrinsic gene signature, and developed an R package named NEPAL, to predict NEPC risk scores. By applying to multiple independent validation datasets, NEPAL consistently and accurately assigned NE feature and delineated PCa progression. Intriguingly, NEPAL demonstrated predictive capabilities for prognosis and therapy responsiveness, as well as the identification of potential epigenetic drivers of NEPC. Conclusion: The present study furnishes a valuable tool for the identification of NEPC and the monitoring of PCa progression through transcriptomic profiles obtained from both bulk and single-cell sources.
Keywords: biomarker; computational biology and bioinformatics; machine learning; neuroendocrine prostate cancer; single-cell.
© The author(s).