Controllable generative AI for peptide design
October 20, 2026 @ 15:15 – 16:15 CEST
Professor Ewa Szczurek, Heimholtz Munich, Germany
Abstract
Antimicrobial resistance is outpacing antibiotic discovery, and antimicrobial peptides offer a promising alternative. AI-driven design requires navigable, controlled and optimized generation from scarce, out-of-distribution activity data. This talk presents a series of models built for that task. HydrAMP, a conditional variational autoencoder, generates prototypes and analogues and yielded 15 validated peptides. PepCompass replaces flat latent search with Riemannian geodesics and Bayesian optimization, reaching a 100% success rate over 25 designs tested against 19 strains. OmegAMP, a guided diffusion model, combines motifs with analogues to control mechanism of action and safety, with 98% of designs active, LPS neutralization dependent on the prototype, and confirmed activity in mice. HyFormer shares weights between prediction and generation in one transformer, with state-of-the-art results on both and a similar experimental hit rate. Work on offline iterative optimization then tackles surrogate inaccuracy and sample inefficiency, with AMPFormer moving designs toward bacterial specificity. Together these results show multi-functional, multi-controllable and experimentally validated design.
Biography
Ewa Szczurek is the director of the Institute of AI for Health at Helmholtz Munich, Germany, and holds professorships at the University of Warsaw (associate), Poland, and LMU Munich, Germany (full), leading joint research laboratories in both cities. She obtained her doctoral degree from the Max Planck Institute for Molecular Genetics in Berlin in 2011, followed by a postdoctoral fellowship at ETH Zurich, and has since held visiting positions at Northwestern University and the Center for Interdisciplinary Research (ZiF) in Bielefeld. Her distinctions include awards from the Rector of the University of Warsaw, IMPRS and ETH Zurich fellowships, a scientific excellence award from the Perspektywy Education Foundation, and recognition as a Forbes Woman to Watch 2026. She serves as an area chair for NeurIPS, ICML and ICLR, and as an associate editor for Genome Biology. Her research develops deep generative models, probabilistic graphical models and representation learning for molecular biology and medicine, with applications in oncology and in AI-driven design of antimicrobial peptides, the subject of her ERC Consolidator Grant.
Host: Vicente Pelechano vicent.pelechano@scilifelab.se
Google Calendar 
iCalendar 
