Fig. 5.

ExplaiNN analysis and interpretation of scATAC-seq data of human pancreatic islets. A Heatmap of the final linear layer weights of an ExplaiNN model with 400 units trained on OCRs from human pancreatic islet sc data [48], with rows representing units with assigned biological annotations based on their Tomtom [27] similarity to known TF profiles from the JASPAR database [18] and columns representing the 12 clusters of cells based on their accessibility profiles and colored by cell type: alpha type 1 cells (purple), beta type 1 cells (navy blue), delta type 1 cells (turquoise), alpha type 2 cells (forest green), beta type 2 cells (light green), delta type 2 cells (yellow), acinar cells (olive green), ductal cell (orange), endothelial cells (red), gamma cells (brown), immune cells (dark brown), and stellate cells (gray). The logos derived from visualizing the filters of selected units are shown at the right. B, C Visualizations of importance scores colored by cell cluster for two units from the previous model that were annotated as RFX and AP1 (TRE site), revealing their respective importance to the prediction of chromatin accessibilities in hormone-high and hormone-low cells. OCR, open chromatin region; scATAC-seq, single-cell ATAC-seq; TF, transcription factor