Development and Implementation of an AWS-Based Platform for Automated Prediction of Antibiotic Resistance from Mass Spectrometry Profiles

Abstract

Antimicrobial resistance (AMR) is a growing threat to global public health, yet few studies have addressed the deployment of intelligent systems for its early prediction in clinical settings. This work presents the development and deployment of a cloud-based web platform that leverages artificial intelligence to predict bacterial resistance using MALDI-TOF mass spectrometry data. The system was developed using a modular architecture deployed on AmazonWeb Services (AWS), combining serverless components and dedicated instances for efficient and scalable operation. A total of 316 clinical isolates of Escherichia coli were collected from the Regional Hospital of Talca, Chile, between 2022 and 2023. A benchmarking analysis comparing CatBoost, XGBoost, and LightGBM was conducted to select the most effective boosting algorithm. The best performance was achieved with Catboost for the Ciprofloxacin case, reaching an AUROC and AUPRC of 0.91. Results for Ceftriaxone were slightly lower, likely due to class imbalance. These outcomes highlight the robustness of boosting models even under real-world data constraints. The entire platform was deployed on Amazon Web Services (AWS) using a modular serverless architecture, enabling scalability, cost-efficiency, and easy integration into hospital workflows. This study demonstrates the feasibility of integrating AI-powered prediction systems into clinical environments to support timely and data-driven antimicrobial resistance management.