Modelling Patient Variation Across Datasets and Diseases With Contrastive Learning... — B.M. Frey

Speaker: B.M. Frey Affiliation: Helmholtz Munich (Institute of Computational Biology) / Technical University of Munich, Germany Date: June 10, 2026 Time: 9:28–9:42am Session: Session 2a – Robust & Multi-modal Learning Recorded at the Helmholtz AI Conference (HAICON26), June 9–11, 2026. More info: https://haicon.cc/conference-program/ About this talk: Modelling Patient Variation Across Datasets and Diseases With Contrastive Learning on Single-Cell Data V.A. Shitov 1, B.M. Frey 1 1 Helmholtz Munich (Institute of Computational Biology) / Technical University of Munich, Germany This work presents SampleCLR, the first batch-aware sample representation method using contrastive multiple-instance learning with intra-dataset hard negative sampling. The method identifies disease trajectories from COVID-19 blood samples and spans blood, gut, and pancreas atlas-scale datasets for age/sex/BMI effects.