Presenter of 1 Presentation
O053 - COMBINATION OF PREDICTIVE CLASSIFICATION ALGORITHMS, USING AUTOMATED TRAINING TECHNIQUES AND MALDI TOF MS FOR CAPSULAR SEROTYPING OF STREPTOCOCCUS PNEUMONIAE. (ID 80)
Abstract
Background
The main aim of this work was to assess if the mass spectra obtained by MALDI-TOF MS, was able to differentiate PCV-13 serotypes from NON-PCV13 serotypes. Therefore using this methodology as screening tool in order to minimize the use of the antisera.
Methods
The training set included PCV-13 isolates and the top 10 of the most prevalent NON-PCV13 isolates, which were selected according to the Argentina national epidemiology, all isolates were previously serotyped by Quellung reaction. Mass spectrum analysis was performed using a MicroFlex LT mass spectrometer (Bruker Daltonik GmbH) and the procedures were conducted according to the manufacturer’s instructions. Classification models were generated using the machine learning (ML) algorithms in ClinProTools, namely QuickClassifier (QC), Supervised Neural Network (SNN), and Genetic Algorithm (GA).
Results
In this first part of the pilot evaluation, we were able to discriminate two groups types, making it possible to differentiate PCV-13 from NON-PCV13 isolates. GA, showed the best cross-validation and recognition capacity values. (Fig1). In the second part, the best models were challenged with 100 isolates whose serotypes were unknown, in order to evaluated the real impact of this approach (Fig2).
Conclusions
A combination of MALDI-TOF MS analysis and ML models may be a potentially efficient screening tool for Streptococcus pneumoniae serotipification, although an external validation must be done in a second part of the pilot evaluation and more isolates whose serotypes are unknown should be challenged with all the algorithms in order to evaluate the real use of this methodology.