Dimension reduction was performed on the DNA accessibility assay dataset using LSI and UMAP as described above. We performed graph-based clustering on the LSI components 2 to 40 by first computing a shared nearest-neighbor graph using the LSI low-dimensional space (with k = 20 neighbors) and then applying the Smart Local Moving algorithm for community detection62. This was performed using the function FindNeighbors with dimensions = 2:40 and reduction = ‘lsi’ followed by FindClusters with algorithm = 3 in Seurat v3.2.0 (ref. 41).
