For each motif (or kmer or genomic annotation), a “raw accessibility deviation” for each cell or sample is computed that represents the difference in the total accessibility of peaks with that motif minus the expected count based on the accessibility profile across all cells, divided by that expected count (Figure S1). Using the matrix of fragment counts in peaks X, where xi,j represents the number of fragments from cell i in peak j, and the matrix of motif matches M, where mk,j is 1 if motif k is present in peak j. The total number of reads mapping to every peak containing motif k in cell i is given by M * XT. For each peak, the expected number of fragments per cell E is computed as the fraction of all fragments across all cells mapping to that peak multiplied by the total number of fragments in peaks for that cell: E=∑i=1xi,j∑j=1∑i=1xi,j∗∑j=1xi,j
