Determining Antibiotic Resistance Utilizing Machine Learning
Abstract
Antimicrobial resistance is a phenomenon observed in infectious disease specialties in which pathogenic microbes become immune to the drugs commonly used to treat infections. The global impact of antimicrobial resistance is such that it recently became only the fourth health topic ever to be discussed at the United Nations General Assembly. In recognition of this problem, the Peer Reviewed Medical Research Program (PRMRP) has identified Antimicrobial Resistance as an important topic area. As rates of drug resistance rise worldwide, there is an increasing need for rapidly identifying bacteria and determining which antibiotics an isolate would be sensitive to. Buried within the genomes of pathogenic bacteria are the genes used to survive antibiotic treatments. Whole genome sequencing (WGS) is becoming cheap, rapid, and can be used to identify these resistance genes. However, the complex data produced by WGS cannot be easily parsed by human judgement. Therefore, to ever be clinically useful as a diagnostic approach, better automated algorithms need to be developed that can accurately predict antimicrobial resistance from WGS data. The PRMRP specifies that diagnostics and treatments targeting multidrug-resistant (MDR) pathogens are a priority for the Discovery Award. This proposal describes an automated pipeline in development that analyzes WGS data from clinically isolated bacteria, identifies mechanisms of resistance, and predicts whether an isolate will be resistant to a variety of currently prescribed antibiotics. This pipeline will eventually enable physicians to make more effective use of antibiotics when treating MDR infections. The WGS analysis pipeline utilizes custom artificial intelligence (AI) algorithms to predict antibiotic resistance in medically important pathogens. A critical requirement for AI training is a sufficient number of examples upon which to learn. This proposal describes the use of publicly available data to train preliminary prediction models and the results achieved thus far. An aim of this proposal is to collect additional data that can be used to improve and externally validate the prediction methods. These methods have also proven to be useful as tools for discovering novel and underappreciated resistance mechanisms. This proposal describes their use to identify new genetic variations that contribute to drug resistance in pathogens, and a proposed approach to validating these findings. This work will result in a significant step forward to realizing the clinical potential of whole genome sequencing as a tool for resistance prediction for bacterial pathogens and will improve the knowledge base concerning drug resistance.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 10, 2021
- Source ID
- W81XWH2010149
Entities
People
- David Greenberg
Organizations
- United States Army
- University of Texas Southwestern Medical Center