Data-Driven Methods for Surveillance and Diagnostics of Antibiotic-Resistant Bacteria

• Antimicrobial resistance is a rapidly growing challenge for the healthcare sector and multi-drug resistant bacterial infections are the cause of nearly a million deaths annually worldwide. Antibiotic resistance is conferred by either antibiotic resistance genes (ARGs) which can be acquired by horizontal gene transfer between bacterial cells or by mutations in pre-existing DNA. Large collections of ARGs are present in the bacterial communities hosted by humans and animals, and in the external environments. Most of these ARGs are uncharacterized and not well-studied. Furthermore, antimicrobial resistance has significantly hindered our ability to treat infections and novel diagnostic solutions are therefore needed to ensure efficient treatment. In paper I, the abundance and diversity of 24,074 ARGs of 17 classes were studied in metagenomic data. The majority of the ARGs were previously uncharacterized, of which several were commonly reoccurring and shared across the digestive system of humans and animals, suggesting that they are under strong selection pressures. The data-driven work in this paper showed that the analysis of all ARGs, including those that have previously not been described, is necessary to provide a comprehensive description of the resistance potential of bacterial communities. In paper II, an AI method for the prediction of bacterial susceptibility towards antibiotics is presented. The method is based on transformers and artificial neural networks and exploits the strong and highly non-trivial dependencies present in the resistance patterns of bacteria. The model was highly successful in predicting susceptibility for most antibiotics from the classes cephalosporins and quinolones but had a lower performance on penicillins and aminoglycosides. The AI-based methodology described in this paper may be used to improve the diagnostics chain of infectious diseases with the potential to reduce the morbidity and mortality of patients. This thesis provides methodologies for improved surveillance and diagnostics of antibiotic-resistant bacteria and, thereby, contributes to a more sustainable use of antibiotics.

Ämnesord

NATURVETENSKAP  -- Biologi -- Bioinformatik och systembiologi (hsv//swe)

NATURAL SCIENCES  -- Biological Sciences -- Bioinformatics and Systems Biology (hsv//eng)

NATURVETENSKAP  -- Matematik -- Annan matematik (hsv//swe)

NATURAL SCIENCES  -- Mathematics -- Other Mathematics (hsv//eng)

Nyckelord

antibiotic resistance

metagenomics

data-driven diagnostics

transformers

Publikations- och innehållstyp

vet (ämneskategori)

lic (ämneskategori)