Fig. 2: Pando leverages multimodal measurements to infer a multiphasic GRN underlying human brain organoid development.

a, Schematic of the Pando GRN-inference framework. Candidate regions are identified through intersection of accessible peaks with CREs or conserved elements. Predicted TFs are selected for each candidate region through binding-motif matching. The relationship between TF–binding-site pairs and the expression of target genes is then fitted with a regression model. E, expression; A, accessibility; g1, target gene 1; tf1,2, transcription factors; p1,4, peaks; GLM, generalized linear model; reg., regularized. b, Signal tracks showing normalized accessibility at the transcription start site of EMX1 in the different branches and inferred regulatory regions for various transcription factors. The line colour represents the sign of the interaction and the box colour (greyscale) represents the false-discovery rate (FDR) of the most significant interaction for this region. c, UMAP embedding of the inferred TF network based on co-expression and inferred interaction strength between TFs. Colour and size represent the expression-weighted pseudotime and PageRank centrality of each TF, respectively. d, UMAP embedding shaded by module features. e, Target specificity for branch-specific TFs. f, UMAP embedding of branch-specific TF networks highlighting TFs with branch-specific targets and interactions with branch-specific accessibility. g, Groups of TFs with differential activity between the dorsal (red) and ventral (purple) telencephalon branch. TF activity is indicated by a coloured dot for each branch, connected by a line, and was calculated by multiplying the mean regulatory coefficient (coef.) with the average expression (expr.) in the branch. The sign of the activity indicates whether the regulation is mainly activating (+) or repressing (−).