Machine learning combined with MALDI-TOF MS has the potential ability to identify serotypes of the avian pathogen Riemerella anatipestifer

Abstract

Aim: Combining MALDI-TOF MS and machine learning to establish a new rapid method to identify two important serotypes of Rimerella anatipestifer. Methods and Results: MALDI-TOF MS was performed on 115 R. anatipestifer strains (serotype 1, serotype 2, and other serotypes) to explore its ability to identify serotypes of R. anatipestifer. Raw spectral data were generated in diagnostic mode; these data were preprocessed, clustered, and analysed using principal component analysis. The results indicated that MALDI-TOF MS completely differentiated serotype 1 from serotype 2 of R. anatipestifer; the potential serotype-associated m/z loci are listed. Furthermore, Random Forest and Support Vector Machine were used for modelling to identify the two important serotypes, and the results of cross-validation indicated that they had ∼80% confidence to make the right classification. Conclusion: We proved that MALDI-TOF MS can differentiate serotype 1 from serotype 2 of R. anatipestifer. Additionally, the identification models established in this study have high confidence to screen out these two important serotypes from other serotypes.