flowVI: Flow Cytometry Variational Inference

Published: 27 Oct 2023, Last Modified: 11 Nov 2023DGM4H NeurIPS 2023 PosterEveryoneRevisionsBibTeX
Keywords: Variational Inference, Deep Generative Models, Multimodal Data Integration, Noise Modeling, Flow cytometry, Feature Imputation, Probabilistic representation, End-to-end Training
TL;DR: flowVI is a deep generative model adept at handling noise variances, technical biases, and batch disparities, seamlessly integrating diverse cytometry datasets and significantly improving protein marker imputation.
Abstract: Single-cell flow cytometry stands as a pivotal instrument in both biomedical research and clinical practice, not only offering invaluable insights into cellular phenotypes and functions but also significantly advancing our understanding of various patient states. However, its potential is often constrained by factors such as technical limitations, noise interference, and batch effects, which complicate comparison between flow cytometry experiments and compromise its overall impact. Recent advances in deep representation learning have demonstrated promise in overcoming similar challenges in related fields, particularly in the context of single-cell transcriptomic sequencing data analysis. Here, we propose flowVI, a multimodal deep generative model, tailored for integrative analysis of multiple massively parallel cytometry datasets from diverse sources. By effectively modeling noise variances, technical biases, and batch-specific heterogeneity using probabilistic data representation, we demonstrate that flowVI not only excels in the imputation of missing protein markers but also seamlessly integrates data from distinct cytometry panels. FlowVI thus emerges as a potent tool for constructing comprehensive flow cytometry atlases and enhancing the precision of flow cytometry data analyses. The source code for replicating these findings is hosted on GitHub, theislab/flowVI.
Submission Number: 43