Desenvolvimento de biossensor para controle de Salmonella Enteritidis baseado em espectroscopia e inteligência artificial associada ou não a biomarcadores ligantes

Abstract

Salmonella spp. is the main pathogen responsible for foodborne illnesses (FADs). Among the serotypes present in the S. enterica species, Salmonella Enteritidis (SE) is primarily associated with human diseases. SE control is mandatory in Brazil and many other countries, but its identification using microbiological tests followed by gold-standard serological and/or molecular tests is time-consuming and laborious for the food industry. His thesis consists of two chapters related to the development of a Fourier Transform Infrared Spectroscopy with Attenuated Total Reflectance (ATR-FTIR) biosensor for the detection of SE in pure colonies and/or bird carcasses, either associated or not with binding biomarkers. The first chapter describes the utilization of Phage Display technology for the selection of SE-binding phages and subsequent detection on the ATR-FTIR biosensor. Following the selection of the top two candidates via Phage-ELISA, the peptides were synthesized and designated as C1-2 and H1-2. These peptides, along with the control antibody (anti-HM), were immobilized on magnetic beads to detect SE in isolated colonies and chicken carcasses. We then utilized an ATR-FTIR biosensor for sample detection, aided by artificial intelligence (AI). The C1-2 peptide exhibited the most promising results, displaying high sensitivity (100%), specificity (91.67%), and an AUC (99.2%) in colonies isolated from SE. For carcass samples, the H1-2 peptide exhibited superior sensitivity (88.57%), specificity (75%), and AUC (76.5%) values compared to C1-2, indicating its potential for implementation in serial tests such as screening. In the second chapter, we combine ATR-FTIR supported by intelligence algorithms to detect SE in pure colony samples. Infrared (IR) spectra were recorded from five Salmonella serotypes [(SE, S. Gallinarum (SG), S. Typhimurium (ST), S. Heidelberg (SH), and S. Dublin (SD)] and the data was randomly split between training data (158 samples) and external validation data (118 samples) to build the Salmonella database. he algorithm models with the best predictive performance were Random Forest, Support Vector Machine (SVM), and Logistic Regression for test data (external validation). All models achieved a high accuracy value of 97.46%. Considering an attractive test to detect Salmonella as one that presents the best sensitivity, specificity, and area under the curve values, the Random Forest model (94.74%, 97.98%, and 99.7% respectively) and SVM (94.74%, 97.98%, and 99.6% respectively) were deemed by this study as the most effective in distinguishing SE from other Salmonella serovars. The dataset from this doctoral thesis suggests that the development of a rapid and sustainable biosensor, supported by artificial intelligence algorithms, along with the selection of molecules using Phage display technology integrated into immunoenzymatic tests, holds significant potential for Salmonella detection.