Influence of culture medium on bacterial molecular profiles in different ionization modes with the use of computational methods

Abstract

Culture medium composition influences the quality of bacterial protein profiles obtained using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS). The study aim was to investigate the influence of three liquid media that differed by carbon source abundance on the differentiation capability of computationally processed mass spectrum (MS) profiles of thirteen human-derived bacteria in positive linear and reflectron (RP) ionization modes. Direct (“intact cell”) and sample extraction protocols were also compared. Unsupervised methods (principal component analysis and hierarchical cluster analysis) allowed observing profile distribution modulated according to the used medium. Optical density results and data analysis approaches showed that Mueller Hinton Broth medium was the most suitable considering both adequate bacterial growth and acquisition of bacteria-specific MS profiles. As it enabled better differentiation among species, extraction was superior to direct (“intact cell”) method. Four machine learning algorithms were tested, aiming at simultaneous classification of bacteria. Random forest model presented the best performance, providing 100% accuracy in multiclass prediction. RP mode analysis provided more diverse MS profiles, displaying also a potential to be used for identification purposes. Additionally, canonical correlation analysis showed that linear and RP modes showed correlation, suggesting that linear mode profiles may be reflected on RP MS patterns.