Abstract

Laboratory surveillance of Streptococcus pneumoniae serotypes is crucial for the successful implementation of vaccines to prevent invasive pneumococcal diseases. The reference method of serotyping is the Quellung reaction, which is labor-intensive and expensive. In the last few years, the introduction of MALDI-TOF MS into the microbiology laboratory has been revolutionary. In brief, this new technology compares protein profiles by generating spectra based on the mass to charge ratio (m/z). We evaluated the performance of MALDI-TOF MS for typing serotypes of S. pneumoniae isolates included in the PCV13 vaccine using a machine learning approach. We challenged our classification algorithms in "real time" with a total of new 100 isolates of S. pneumoniae from Argentinian nationwide surveillance. Our best approach could correctly identify the isolates with a sensitivity of 58.33 % ([95%IC 40.7-71.7]); specificity of 81.48 % ([95%IC 53.6-79.7]); accuracy of 63.0% ([95%IC 61.9-93.7]); PPV of 80.77% ([95%IC 64.5-90.6]) and NPV of 59.46% ([95%IC 48.9-69.2]). In this work, it was possible to demonstrate that the combination of MALDI-TOF mass spectrometry and multivariate analysis allows the development of new strategies for the identification and characterization of Spn isolates of clinical importance; and we consider that by using AI, as more data becomes available the models will get better and more precise.