Diagnóstico do câncer de próstata utilizando metabolômica exploratória e ATR-FITR: prova de conceito

Abstract

Introduction: The diagnosis of prostate cancer faces several challenges, one of the main problems is the absence of a specific cancer marker and the complications of prostate biopsy. These challenges highlight the need for more disruptive and effective diagnostic tools to improve diagnostic accuracy. Objective: To identify metabolites and develop diagnostic platforms using exploratory metabolomics and ATR-FTIR with artificial intelligence algorithms for the diagnosis of prostate cancer. Methodology: In exploratory metabolomics, 46 participants were divided into 3 groups: group 1 (patients diagnosed with prostate cancer n = 19), group 2 (postoperative radical prostatectomy n = 11), group 3 (control n = 16), from which blood samples were collected. Metabolites were extracted with methanol, the data obtained were analyzed with an exploratory metabolomics strategy and creation of algorithms by intelligence to classify the groups. In ATR-FTIR, 20 participants were divided into 2 groups: group 1 (patients diagnosed with prostate cancer n = 11), group 2 (control n = 9), from which blood samples were collected. Serum spectral data were obtained using the Agilent Cary 630 FTIR Spectrometer in microattenuated total reflectance (ATR) mode, covering a wavenumber range from 4000 to 650 cm⁻¹. Spectral data were analyzed in AI tools. Results: In exploratory metabolomics, a total of 1,744 metabolites were identified. A 100% frequency filter was used, ANOVA analysis with p < 0.05 and fold change greater than or equal to 2 and 11 differentially expressed metabolites were obtained. Among these results with differences, 2 metabolites stand out for being absent in group 1: Pi22 refers to a certain phosphoinositide (PI) and C19 Sphingosine-1-phosphate. Using artificial intelligence, an algorithm was created in the Gradient Boosting model with 96% sensitivity and 98% specificity and 97% accuracy for the diagnosis of prostate cancer. Using artificial intelligence (AI) tools, the results demonstrated that the Neural Network algorithm provided the most effective discrimination between patients in the 2 groups. This preprocessing approach with 100% sensitivity, 100% specificity and 100% accuracy in differentiating between the groups. Conclusion: Exploratory metabolomics evaluations of serum and ATR-FTIR of serum combined with machine learning present a potential diagnostic tool for prostate cancer. The use of exploratory metabolomics found promising data for two metabolites that were absent in group 1, and are indicated for future studies as possible biomarkers. It was also possible to develop an artificial intelligence algorithm with high sensitivity, specificity and accuracy for the diagnosis of prostate cancer.