MACHINE LEARNING TOOLS IDENTIFY A P NOSTIC SIGNATURE IN PROSTATE CANCER THAT OUTPERFORMS CURRENT RISK PREDICTORS

BIZZOTTO, JUAN; SABATER, AGUSTINA; LAGE-VICKERS, SOFIA; SANCHIS, PABLO; VAZQUEZ ELBA,; VALACCO, MP; COTIGNOLA, JAVIER; GUERON GERALDINE

Mar del Plata

Congreso; REUNIÓN CONJUNTA SAIC SAB AAFE AACYTAL 2023 LXVIII REUNIÓN ANUAL DE LA SOCIEDAD ARGENTINA DE INVESTIGACIÓN CLÍNICA (SAIC); 2023

Sociedad Argentina de Investigación Clínica

In Argentina, prostate cancer (PCa) is the most common cancer in men. Based on histopathological characteristics, some PCa patients are grouped as intermediate risk, but they can vary significantly in the outcome of the disease. Our aim was to identify molecular biomarkers that could stratify the risk of progression in PCa patients, independent of histological grading and other clinicopathological variables. In this study, we performed protein extraction from formalin-fixed paraffin-embedded PCa and benign prostatic hyperplasia (BPH) tissue samples. Subsequently, tandem mass spectrometry (LC ESI-MS/MS) analysis enabled the identification of 109 proteins enriched in PCa compared to BPH samples. We then subjected these proteins to integrated bioinformatics analysis using publicly available transcriptomic databases (16 datasets, n = 2.954). For this purpose, we developed a sequential workflow including differential expression analysis, survival analysis, and identification of key predictors of PCa progression using machine learning techniques. Our results identified a gene expression signature capable of identifying patients at higher risk of progression of the disease, independent of the evaluated clinicopathological parameters, outperforming histological Gleason grading (GG) in the intermediate-risk patient subgroup. Further, we validated these findings in new PCa datasets that were not used during the training phase, underscoring the robustness of our methodology. In summary, leveraging experimental data, we established a workflow for transcriptomic data analysis and developed a gene signature outperforming GG risk prediction, even among patients previously classified as intermediate risk. This signature will enable more precise early diagnosis, facilitating personalized treatment, improving clinical outcomes, and reducing unnecessary interventions.