Abstract 3388: Proteomic analysis of Latin American patients and machine learning strategies for developing a prognostic signature for prostate cancer

Abstract Prostate cancer (PCa) is the leading cause of cancer-related deaths among males in Latin America (LATAM). While some PCa patients are classified as intermediate-risk based on histopathological characteristics, their outcomes can vary widely. There is an unmet need for biomarkers that are cost-effective, easy to implement in low-resource clinical settings, and capable of identifying patients at risk for aggressive disease to tailor treatments. These biomarkers must be developed specifically for the LATAM population to ensure their relevance and accuracy within this demographic. This study aimed to identify molecular biomarkers capable of stratifying the risk of disease progression in LATAM PCa patients, independent of histological grading and other clinicopathological variables. To this end, we performed protein extraction from formalin-fixed paraffin-embedded (FFPE) PCa and benign prostatic hyperplasia (BPH) tissue samples from patients at Buenos Aires, Argentina (n=20). Tandem mass spectrometry (LC ESI-MS/MS) analysis identified 109 proteins enriched in PCa. These proteins were subsequently analyzed through a custom made bioinformatic pipeline using publicly available transcriptomic datasets (16 datasets, n = 2, 954). We designed a sequential workflow that included survival analyses to evaluate the association of each gene with disease progression, as well as machine learning algorithms to identify key predictors of PCa progression. Random forest was used for feature selection, identifying the most important genes, and LASSO regression was applied to construct a predictive model based on these genes. Our analysis yielded a 10-gene expression signature capable of identifying patients at higher risk of disease progression, independent of clinicopathological parameters, specifically within the intermediate-risk subgroup (HR > 1, p < 0.05 across datasets). We validated these findings in 4 independent PCa datasets not used in the training phase, reinforcing the robustness and generalizability of our methodology. This study leveraged experimental and transcriptomic data to develop a prognostic gene-expression signature based on genes with confirmed expression and detectability in the LATAM population, ensuring its relevance and applicability within this demographic. The signature has the potential to enable risk assessment among intermediate-risk patients, facilitating personalized treatment strategies, improving clinical outcomes, and reducing unnecessary interventions. Additionally, the small size of the gene panel makes it cost-effective and compatible with standard RT-qPCR platforms, enabling widespread implementation in decentralized settings. Our results lay the foundation for future clinical trials integrating this signature into personalized therapeutic approaches. Citation Format: Juan Bizzotto, Agustina Sabater, Pablo Sanchis, Pia Valacco, Elba Vazquez, Javier Cotignola, Geraldine Gueron. Proteomic analysis of Latin American patients and machine learning strategies for developing a prognostic signature for prostate cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3388.