MicroBIS - An integrated AI laboratory Assistant for Bacteria and Antimicrobial Resistance Identification

Antimicrobial Resistance (AMR) represents a major global health challenge, placing considerable strain on both clinical and public health systems—a burden projected to escalate without substantial intervention. Surveillance programs like the Pfizer Antimicrobial Testing Leadership and Surveillance (ATLAS) initiative play a critical role in addressing this challenge by providing valuable data to advance AMR research. In clinical settings, machine learning (ML)-based, data-driven tools have been proposed to aid bacterial identification and antibiotic prescription decision-making, aiming to minimize the human error inherent in these processes. In this work, we integrate the Pfizer-ATLAS dataset into MicroBIS, a novel ML-based platform for bacterial identification, and antibiotic susceptibility profiling. MicroBIS employs a Random Forest model trained on both simulated biochemical test results and real-world data obtained from Pfizer ATLAS program. Our results show that MicroBIS achieves a 79% accuracy rate in predicting bacterial species from 20-panel biochemical tests and a 31% accuracy from 8-panel biochemical tests. When identifying the top 5 bacterial species out of 133 candidates, MicroBIS achieved accuracies of 99% and 74% for the 20-panel and 8-panel biochemical tests, respectively. Additionally, MicroBIS predicted antimicrobial susceptibility profiles with an overall accuracy of 56% using the Pfizer ATLAS dataset. We developed MicroBIS as an integrated pipeline that combines bacterial identification with antibiotic susceptibility prediction, providing a clinically oriented platform that leverages simple laboratory inputs. MicroBIS offers significant potential to enhance clinical decision-making, bolster public health surveillance, and support AMR research efforts, especially in resource-limited settings.