Current students

New large-scale studies are now providing detailed information about individual cells from many people. We also have powerful foundational models available that have learned from existing collections of cell data. This combination gives us the opportunity to build a new way to predict and understand diseases from single-cell transcriptomics. ; We will create a framework that uses Multiple Instance Learning (MIL) models to both probabilistically classify donor single-cell profiles according to their disease status or biomarker abundance and simultaneously identify implicated cell types and/or states. To do this, we will combine both single-cell foundation models and MIL methodologies, marking a paradigm shift in enabling disease prediction and a single-cell level understanding of disease heterogeneity. Specifically, we will leverage single-cell RNA-seq profiles collected in 5000 densely phenotyped UK biobank donors. We will reach a characterization of how cell states are implicated in specific diseases, their progression, and, possibly, metabolic biomarkers.