We next characterized chromVAR’s ability to capture biologically relevant chromatin variability from single cell ATAC-seq data drawn from multiple distinct cell lines and human samples (Supplementary Fig. 8). Using tSNE with bias corrected deviations for motifs and 7mers, we clustered individual cells into distinct cell types and observe individual motifs that best distinguish each cell type (Fig. 2a). Notably, well defined, distinct clusters are formed in this tSNE projection when using the bias corrected deviations, but the clustering is confounded by technical biases when using raw deviations without the bias correction infrastructure. Importantly this approach for classifying cell types also compares favorably performing tSNE on the counts within peaks using a variety of approaches (Supplementary Fig. 9). Interestingly, we also observe that cells from acute myeloid leukemia (AML) patients cluster between lymphoid-primed multipotent progenitors (LMPPs), monocytes, and HL60 (an AML derived cancer cell line) cells. In this unsupervised analysis, we find that the AML leukemic stem cells are more similar to LMPPs, while the AML blasts are more similar to the monocytes. In addition, we also observe that patient 1
