Figure A: This table denotes the constituent motifs across all tasks (i.e. cell types) of this TF, which constitute an aggregate motif identity. These motifs are merged across the count and profile heads for each model. Here, we show the CWMs.

Figure B: For each motif discovered across all cell types, we show the CWM. The presence of a blue circle denotes that the motif was found in that cell line. The area of the circle reflects the proportion of peaks that contain the motif, with a perfectly inscribed circle representing all peaks. For the set of peaks in each cell line, we show how closely MEME, HOMER, or DiChIPMunk is able to recover the same motif, via cosine similarity of the closest motif to the TF-MoDISco motif. We also show the closest JASPAR match to the TF-MoDISco motif.