Detection of Mycobacterium avium ssp. paratuberculosis in Cultures From Fecal and Tissue Samples Using VOC Analysis and Machine Learning Tools

• DOI: 10.3389/fvets.2021.620327
• Publication Date: 2021-02-03T06:30:49Z
• Authors: Philipp Vitense; Elisa Kasbohm; Anne Klassen; Peter Gierschner; Phillip Trefz; Michael Weber; Wolfram Miekisch; Jochen K. Schubert; Petra Möbius; Petra Reinhold; Volkmar Liebscher; Heike Köhler
• Keywords: random forests; 2. Zero hunger; 0303 health sciences; 03 medical and health sciences; machine learning; paratuberculosis; bacterial culture; Veterinary medicine; SF600-1100; diagnostics; Veterinary Science; Mycobacterium avium ssp. paratuberculosis

Analysis of volatile organic compounds (VOCs) is a novel approach to accelerate bacterial culture diagnostics Mycobacterium avium subsp. paratuberculosis (MAP). In the present study, cultures fecal and tissue samples from MAP-infected non-suspect dairy cattle goats were explored elucidate effects sample matrix animal species on VOC emissions during cultivation identify early markers for growth. The processed following standard laboratory procedures, tubes incubated different time periods. Headspace volume was sampled by needle trap-micro-extraction, analyzed gas chromatography-mass spectrometry. MAP-specific considered potential characteristic patterns. To address variation patterns, flexible robust machine learning workflow set up, based random forest classifiers, comprising three steps: variable selection, parameter optimization, classification. Only few substances originated either certain or could be assigned one species. These additional not informative selection procedure. Classification accuracy MAP-positive negative bovine feces 0.98 caprine 0.88, respectively. Six indicating MAP presence selected in all four settings (cattle vs. goat, tissue): 2-Methyl-1-propanol, 2-methyl-1-butanol, 3-methyl-1-butanol, heptanal, isoprene, 2-heptanone. accuracies growth-scores ranged 0.82 goat 0.89 feces. Misclassification occurred predominantly between related scores. Seventeen growth settings, including 6 presence. concentration levels 2,3,5-trimethylfuran, 2-pentylfuran, 1-propanol, 1-hexanol indicative before visible apparent. Thus, very accurate classification achieved analysis detect colonies become confirmed. results indicate that diagnosis can optimized monitoring cultures. Further validation studies are needed increase robustness patterns as pre-requisite development VOC-based diagnostic systems.