Multi-Pathogen Antimicrobial Resistance Prediction Using Supervised and Deep Tabular Learning on the Pfizer ATLAS Global Surveillance Dataset

This project aims to develop and benchmark machine learning models for predicting antimicrobial resistance (AMR) phenotypes across WHO-priority bacterial pathogens using the Pfizer ATLAS dataset. The study will leverage patient demographic data, infection source, geographic metadata, and minimum inhibitory concentration (MIC) values to train classification models (XGBoost, LightGBM, TabNet, and deep learning architectures) for resistance outcome prediction.
Secondary objectives include spatiotemporal trend modeling of resistance spread across regions and years, multi-label drug susceptibility prediction across antibiotic classes, and detection of anomalous resistance patterns indicative of emerging threats. The dataset will be used strictly for academic, non-commercial research purposes. No patient re-identification will be attempted. Data will be stored securely and accessed only by the research team. Results will be published openly to contribute to the global AMR research community.