Abstract
Streptococcus pneumoniae is the leading cause of community-acquired pneumonia and remains a significant contributor to bacteremia and meningitis, collectively known as invasive pneumococcal disease (IPD). Certain serotypes are more strongly associated with severe illness and antimicrobial resistance. Accurate serotyping is essential for effective IPD surveillance and vaccine development. Fourier-transform infrared (FTIR) spectroscopy has emerged as a valuable tool for differentiating among serotypes across various isolates. We analyzed 150 pneumococcal strains isolated from a tertiary hospital in Barcelona, Catalonia, Spain, between 2016 and 2023, representing 32 serotypes associated with IPD. Forty-nine samples (33%) exhibited serotypes included in PCV13 vaccine. Each strain was classified using (A) FTIR-based clustering and (B) FTIR machine-learning-based PneumoClassifier algorithm. The results were compared to the Quellung reaction, the gold standard methodology. Clustering method grouped correctly PCV13-serotypes 1, 3, and 19F and non-PCV13 serotypes 6C, 7BC, 17F, 24F, 31, and 35B (48/150). PneumoClasifier algorithm successfully grouped all PCV13-serotypes (49/49) including some of the most virulent described serotypes, such as 1, 6B, 7F, and 14. Among non-PCV13 serotypes, it correctly classified 73 out of 101 isolates (72.3%). However, 12F, 15AB, 16F, 17F, 23A, and 24F were misclassified. Overall, PneumoClassifier achieved an accuracy of 122/150 (79.80%) in serotyping pneumococcal strains, demonstrating higher concordance with Quellung (adjusted Rand index: 0.717, adjusted Wallace coefficient: 0.636) compared to the clustering approach (0.397 and 0.378, respectively) (p < 0.001). FTIR has proven to be a rapid, user-friendly, cost-effective, and practical technique, making it a promising first-line tool for S. pneumoniae serotyping.