SNIDDY Seminar Series: AMR and AI

Our next online seminar will feature Diane Duroux on Friday, March 13th 2026, 1–2pm

Title: Early and Personalized Detection of Antimicrobial Resistance Using Mass Spectrometry and AI

Abstract: Patients with suspected infections often receive empiric antibiotics while undergoing a two-stage diagnostic process. Pathogens are first identified at the species level within 24–48 hours using MALDI-TOF mass spectrometry, followed by antimicrobial susceptibility testing after an additional 24–48 hours to determine resistance profiles. We aim to bridge the gap between speed and precision by developing AI-assisted tools to identify antimicrobial resistance directly from MALDI-TOF spectra. We advance prior work by developing multimodal models that integrate proteomic spectra with chemical representations of antimicrobials, improving predictive performance compared to traditional drug–species models. In addition, to address the common problem of poor generalization in clinical AI, we learn robust feature representations of spectra. Evaluated across four institutions, our models show improved performance and stability across hospitals and time periods. This framework supports earlier optimization of antibiotic therapy by improving initial drug selection and enabling faster de-escalation to effective narrow-spectrum treatments.