Fig. 1

Overview of the scover workflow. a Cell pooling strategy. A subset of cells is selected using geometric sketching. The k-nearest neighbors for cells in the subset are summed to produce a less sparse representation of the original dataset. b Scover infers regulatory motifs that are predictive of the signal associated with a set of sequences using a neural network consisting of a single convolutional layer, an exponential linear unit, global max pooling, and a linear layer with bias term. c The identified motifs are merged and compared to annotated motifs and assigned to pre-specified motif clusters based on their most significant alignment. d To evaluate the contribution of each motif, an influence score is calculated using a leave-one-out strategy