Re-visiting clonal-complex classifications: a novel machine learning approach for investigating population dynamics of antimicrobial resistance in Campylobacter jejuni

Thesis

Campylobacter jejuni (C.jejuni) is a widespread and highly diverse gram-negative zoonotic bacterium, which is a leading cause of bacterial food poisoning worldwide. Robust systems of variant classification, using Multi-locus Sequence Typing (MLST), remain a vital part of understanding the epidemiologic features of C. jejuni. This work re- evaluates clonal-complex classification and identifies inconsistencies in specific variants, such as clonal-complex 353, among modern whole-genome sequenced isolates. In addition, this thesis introduces a novel approach to MLST classification called “NeighbourGroups”, a generalisable machine learning approach that reproducibly, robustly, and rapidly classifies 7 housekeeping MLST loci onto core genome level groupings. This work has determined that for C.jejuni, we need 20 variant groups; we undertook a comparative analysis of each isolate between the two classification: clonal- complex and NeighbourGroups. This work identified sub-variant grouping among CC21 and CC353 and revealed novel dynamics of antimicrobial resistant among C.jejuni. Furthermore, potential new housekeeping genes have been identified through this study indicating that this same approach could be applied to other pathogens to identify one locus that can demonstrate cgMLST level NeighbourGroups. This approach could be then applied across pathogens to further identify potential AMR targeted locus that we have not currently sought for. Moreover, once we have locus per pathogen we could even identify if we do have a housekeeping gene across pathogens that could be used for speciation. Furthermore, current clonal-complex schemes have been re-evaluated to show if we want to keep using it which one can stay the same, which one should be changed, and which ones could be retired as a clonal-complex due to lack of isolate counts.

In summary, this thesis has both re-evaluated the gold standard of molecular epidemiology, identified issues that shown it cannot be resolved in clonal-complex level and created a solution through the establishment of NeighbourGroups which will both support the newly whole genome sequenced isolates and past historical isolates that we only have MLST information. It is hoped that the methods described will be applied beyond C.jejuni and contribute positively to epidemiological research.

Authors: Veltcheva, DL

• DOI: 10.5287/ora-pmx5zamxm
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: antimicrobial resistance; Campylobacter Jejuni; antibiotic resistance; whole-genome sequence; MLST; epidemiology; zoonotic; cgMLST; machine learning; AMR; pathogen; clonal-complex; multi-locus sequence typing
• Subjects: Statistics; Microbiology; Biology; Bioinformatics; Machine learning; Molecular epidemiology