ConvexGating infers gating strategies from clusters in single cell cytometry data

Abstract

Manual expert gating remains common practice for defining specific cell populations in flow cytometry data, but increasing numbers of measured parameters and high inter-rater variability limit consistency across studies. Cluster-based approaches use the full marker space to define cell populations more consistently, but their outputs cannot be directly implemented on a cell sorter. Here we develop ConvexGating, an artificial intelligence tool to address this gap by automatically learning interpretable gating strategies for sorting in an unbiased, data-driven manner, generating low-contamination strategies for both known and previously unknown cell populations, including plasmacytoid dendritic cells identified solely as CD57-CD13-CD45RA+ CD123+ cells. We show that ConvexGating derives sorting strategies for CD8+ subtypes and adipose progenitor cell populations, which we validate experimentally by single-cell sequencing of sorted cells. We also demonstrate that the method transfers effectively to Cytometry by Time of Flight and Cellular Indexing of Transcriptomes and Epitopes by Sequencing data and improves marker panel design for cell sorting. Here, the authors develop ConvexGating, an AI tool that learns interpretable, data-driven gating strategies for cell sorting. They show that when analyzing scRNA-seq data, it yields low-contamination populations and also works across flow cytometry, cyTOF and CITE-seq data.