Fig. 2: Mutual information regularization disentangles the influence of biological and technical effects.
From: Multi-batch single-cell comparative atlas construction by deep learning disentanglement
a Predicted gene expression rates versus predicted technical effects for naïve B-cell, NK/CD8+ T-cell, and monocyte marker genes estimated using a topic model trained with the marginal likelihood maximization objective (\({{{{{{\mathcal{V}}}}}}}_{{{{{{\rm{ELBO}}}}}}}\)). Colored by observed expression counts, expert-annotated cell types, and batches. The batches are labeled according to single-cell assay site (s1-s4) and sample donor (d1-d10). b Predicted expression rates versus technical effects for the same genes, estimated using a topic model trained with mutual information regularization (\({{{{{{\mathcal{V}}}}}}}_{{{{{{\rm{CODAL}}}}}}}\)). Source data are provided as a Source Data file.
