scANVI: an extension to scVI for semi‐supervised annotation

scVI is a hierarchical Bayesian model (Gelman & Hill, 2006) for single‐cell RNA sequencing data with conditional distributions parametrized by neural networks. The graphical model of scVI (Fig 1C) is designed to disentangle technical signal (i.e., library size discrepancies, batch effects) and biological signal. We propose in this manuscript an extension of the scVI model to include information about cell types in the generative model. We name this extension scANVI (single‐cell ANnotation using Variational Inference).