Identification of a 9-gene signature to enhance biochemical recurrence prediction in primary prostate cancer: A benchmarking study using ten machine learning methods and twelve patient cohorts

• Published in: Cancer letters 2024-04, Vol.588, p.216739-216739, Article 216739
• Main Authors: Yin, Wenjun, Chen, Guo, Li, Yutong, Li, Ruidong, Jia, Zhenyu, Zhong, Chuanfan, Wang, Shuo, Mao, Xiangming, Cai, Zhouda, Deng, Junhong, Zhong, Weide, Pan, Bin, Lu, Jianming
• Format: Article
• Language: English
• Subjects: Biochemical recurrence; Machine learning; Prostate cancer; Radical prostatectomy
• ISSN: 0304-3835; 1872-7980
• DOI: 10.1016/j.canlet.2024.216739

Prostate cancer (PCa) is a prevalent malignancy among men worldwide, and biochemical recurrence (BCR) after radical prostatectomy (RP) is a critical turning point commonly used to guide the development of treatment strategies for primary PCa. However, the clinical parameters currently in use are inadequate for precise risk stratification and informing treatment choice. To address this issue, we conducted a study that collected transcriptomic data and clinical information from 1662 primary PCa patients across 12 multicenter cohorts globally. We leveraged 101 algorithm combinations that consisted of 10 machine learning methods to develop and validate a 9-gene signature, named BCR SCR, for predicting the risk of BCR after RP. Our results demonstrated that BCR SCR generally outperformed 102 published prognostic signatures. We further established the clinical significance of these nine genes in PCa progression at the protein level through immunohistochemistry on Tissue Microarray (TMA). Moreover, our data showed that patients with higher BCR SCR tended to have higher rates of BCR and distant metastasis after radical radiotherapy. Through drug target prediction analysis, we identified nine potential therapeutic agents for patients with high BCR SCR. In conclusion, the newly developed BCR SCR has significant translational potential in accurately stratifying the risk of patients who undergo RP, monitoring treatment courses, and developing new therapies for the disease. •We have collected and curated global multicenter primary prostate cancer RNA sequencing data and follow-up data from 12 centers including Europe, America and Asia, and made them available for subsequent researchers to use.•A computational framework was included, consisting of 101 machine learning algorithm combinations, to identify a novel signature called BCR SCR for predicting the risk of biochemical recurrence after radical prostatectomy.•BCR SCR demonstrated superior performance compared to 102 previously established models for prostate cancer.•BCR SCR excels in forecasting post-surgery prostate biochemical recurrence, outdoing 102 models.