CytoVI: deep generative modeling of antibody-based single cell data.

CytoVI accurately models antibody-based single-cell data, revealing proteins linked to immunoglobulin class-switching and disease-associated T cell states in B cell non-Hodgkin lymphoma.

• Why it matters: Analyzing antibody-based single-cell data is hindered by technical noise, batch effects, platform differences, and limited antibody panels, which impede reliable biological insights.
• What they did: CytoVI employs a probabilistic generative approach to generate cell embeddings, impute missing data, perform differential expression testing, and automate cell annotation, applied to a dataset of 350 proteins.
• The result: The model identified proteins associated with immunoglobulin class-switching and uncovered disease-related T cell states, enabling more accurate and integrated analysis of complex single-cell datasets.