The advent of multimodal single cell data has opened new avenues to explore single cells biology. Here we explore the application of a Latent Dirichlet Allocation (LDA) model to multimodal single cell data. We see that naive horizontal integration is not viable, at least with LDA. However, we do see strong results when LDA is applied to multimodal, scRNA and scATAC, 10X Multiome data. We notice that lower dimensional representation of the topic model can accurately separate cell types. Finally, we also see that the topics themselves seem to capture biologically relevant information and, in the future, could be used to learn the underlying pathways which lead to immune cell activation.
Class project: UCLA STATS254 Winter'22