Figure 7.
Jointly clustering and annotating cells. The autoencoder contains two parts: the cluster assignment part (gray) uses a low-dimensional representation, ze, to assign a cell to clusters; the “annotator” (green) uses the learned gene set activity scores and selected genes’ expression to refine clustering and annotate clusters. Gene sets and genes selected as predictive by the annotator, in turn, provide useful annotations for each cell cluster. We set the number of clusters M as six in this figure for illustration purposes.
