In addition to visualizing the similarity of cells, we inverted our tSNE analysis to visualize the similarity of motifs and kmers in their activity patterns across cells (Fig. 3a). In this visualization, motifs and kmers that have similar activity profiles across cells cluster together in the tSNE subspace, allowing the identification of major clusters representing several different TF families. Notably, different TFs within the same family (e.g. GATA1 and GATA2) often bind highly similar motifs, and therefore chromVAR alone cannot distinguish the causative regulator binding a particular TF motif. In the inverted tSNE visualization for motif and kmer similarity, most, but not all kmers cluster with a known motifs, suggesting k-mer analysis may enable de novo discovery of previously unannotated motifs.
