Machine Learning for AMR prediction in African Pathogens

This study aims to develop and validate machine learning models for predicting antimicrobial resistance in priority bacterial pathogens affecting African populations. Specific objectives include:
Primary Objective:
Develop machine learning classifiers to predict antimicrobial resistance profiles from bacterial genomic sequences for E. coli and K. pneumoniae against first-line and last-resort antibiotics
Secondary Objectives:
Identify genomic features and resistance mechanisms most predictive of phenotypic resistance
Evaluate model generalizability across different geographic regions, with emphasis on African strain data
Compare performance of different ML architectures (graph neural networks, gradient boosting, ensemble methods)
Develop interpretable models that identify resistance genes and mechanisms
Tertiary Objective:
Create a proof-of-concept web-based tool for clinicians to input genomic data and receive resistance predictions

Clinical Significance
Rapid diagnostics: Genomic sequencing can be completed in 4-6 hours, and ML prediction in seconds, compared to 24-72 hours for culture-based testing
Improved treatment decisions: Accurate resistance prediction enables targeted antibiotic selection, reducing inappropriate use
Outbreak response: Rapid identification of resistance patterns aids infection control measures

Scientific Innovation
African-focused validation: Most AMR prediction models are trained predominantly on Western data; this study will specifically validate performance on African bacterial strains
Multi-drug prediction: Simultaneous prediction of resistance to multiple antibiotics from a single genomic input
Mechanism discovery: Interpretable ML approaches to identify novel or emerging resistance mechanisms
Integration-ready: Design for potential integration with existing genomic surveillance systems

Public Health Impact
Addresses the highest AMR burden globally (sub-Saharan Africa)
Supports antimicrobial stewardship efforts in resource-limited settings
Provides actionable data for targeted interventions against priority pathogens