Adaption of the Aristotle Classifier for Accurately Identifying Highly Similar Bacteria Analyzed by MALDI-TOF MS

Abstract

MALDI-TOF MS has shown great utility for rapidly identifying microbial species. It can be used to successfully type bacteria and fungi from a variety of sources more rapidly and cost-effectively than traditional methods. One area where improvements are necessary is in the typing of highly similar samples, such as those samples from the same genus but different species or samples from within a single species but from different strains. One promising way to address this current limitation is by using advanced machine learning techniques. In this work, we adapt a newly developed machine learning tool, the Aristotle Classifier, to bacterial classification of MALDI-TOF MS data. This tool was originally developed for classifying glycomics and glycoproteomics data, so we modified it to be well-suited for assigning mass spectral data from bacterial proteins. The classifier exceeds existing benchmarks in classifying bacteria, and it shows particularly strong performance when the samples to be identified are highly similar. The combination of mass spectrometry data and tools like the Aristotle Classifier could ameliorate the ambiguities associated with challenging bacterial classification problems.

Machine learning for microbial identification and antimicrobial susceptibility testing on MALDI-TOF mass spectra: a systematic review

“…Lastly, the reproducibility of the reviewed studies is low. MALDI-TOF MS datasets (in our review only nine studies [25,28,37,38,46e48,50,53]) and machine learning codes (in our review only five studies [12,34,37,50]) are rarely made public after publication. Additionally, the information on hyperparameter choices for models is often insufficient [14,15,26,28e31,33,35,40,51].…”

Recent evolutions of machine learning applications in clinical laboratory medicine

“…While this approach is common practice in the routine clinical microbiology lab, a lot of useful information remains, unfortunately, unexploited [89]. Therefore, several research groups have been applying ML algorithms to fully exploit the valuable information contained in MALDI-TOF MS spectra [89][90][91][92][93][94][95][96]. As an example, ML models (e.g.…”

“…SVM, RF, and ANN) have been recently developed to allow rapid classification of group B Streptococcus serotypes [90], to differentiate between Escherichia coli and Shigella species [91], to strain type Staphylococcus haemolyticus [93], to discriminate between Clostridium species [94], and to differentiate between different Klebsiella species [95]. Furthermore, Desaire and Hua [92] adapted an existing ML tool, originally developed for classifying glycomics and glycoproteomics data, to accurately distinguish between highly similar bacteria analyzed by MALDI-TOF MS. Overall, the authors reported an outperformance of the model compared to existing benchmarks.…”

Clasificadores de Aprendizaje Automático aplicados a Datos RKI MALDI-TOF de Espectrometría de Masa de Bacterias

“…Mientras que la secuenciación y los métodos convencionales de pruebas bioquímicas y determinación del fenotipo aplicados a la identificación de patógenos son costosos en dinero y tiempo (Zhu et al, 2015), MALDI-TOF/MS surge como una alternativa rápida, robusta y más económica. Trabajos recientes han demostrado el interés en el empleo de este método (Lasch et al, 2016;Avanzi et al, 2017;Manukumar y Umesha, 2017;Desaire y Hua, 2019). En los últimos años, MALDI-TOF/MS se ha convertido en la herramienta de identificación de primera línea de microorganismos.…”