| Abstrakt: |
Invasive pneumococcal disease (IPD) is a major cause of morbidity and mortality worldwide, particularly in the pediatric population (children and infants), with high rates of hospitalization and death. This study aimed to create and validate a classifier for Streptococcus pneumoniae serotyping using Fourier-transform infrared (FT-IR) spectroscopy as a rapid alternative to the classical serotyping technique. In this study, a database comprising 76 clinical isolates, including 18 serotypes (predominantly serotypes 19A, 6C, and 3) of S. pneumoniae from pediatric patients with IPD, was tested at a tertiary pediatric hospital in southern Brazil during 2016–2023. All isolates were previously serotyped using the Quellung reaction, and 843 FT-IR spectra were obtained to create a classification model using artificial neural network (ANN) machine learning. After the creation of this classifier, internal validation was performed using 384 spectra as the training dataset and 459 as the testing dataset, resulting in a predictive accuracy of 98% for serotypes 19A, 6, 3, 14, 18C, 22F, 23A, 23B, 33F, 35B, and 9N. In this dataset, serotypes 10A/16F, 15ABC, and 7CF could not be differentiated and were, therefore, grouped as labels. FT-IR is a promising, rapid, and low-cost method for the phenotypic classification of S. pneumoniae capsular serotypes. This methodology has significant implications for clinical and epidemiological practice, improving patient management, monitoring infection trends, and developing new vaccines. [ABSTRACT FROM AUTHOR] |
