Fig. 3: Network-level convergence resolves cell-type-specific and developmental-specific node genes.
‘Convergent networks’ are co-expressed genes that share similar expression patterns across NDD gene perturbations, here resolved for the nine NDD KOs resolved across all four cell types: ARID1B, ASH1L, CHD2, MED13L, NRXN1, PHF21A, SETD5, SIN3A, SMARCC2. a, Schematic explaining cell-type-specific convergence at the network level using Bayesian bi-clustering and unsupervised network reconstruction. b, Strength of network convergence across all random combinations of nine NDD KO perturbations by cell type. (i) The mean strength of network convergence is significantly different by cell type, with the highest convergence present in immature iGLUTs. The same KO combinations tested in one cell type may not resolve convergence in another cell type. Each point represents a resolved network, and its calculated convergence strength. Dots that represent the same combinations of KO perturbations, but tested in each cell type, are connected by a line. In the box plots, the median is represented by the line (center) and the mean as the red point. The lower and upper hinges correspond to the first and third quartiles (the 25th and 75th percentiles). The upper and lower whiskers extend from the hinge to the largest or smallest value up to 1.5 × IQR. All data points are plotted individually. c, Convergent network strength was most correlated between mature iGLUTs and iGABAs (two-tailed Pearson’s correlation test with Holm’s multiple testing correction; PCC = 0.6, PHolm < 2.2 × 10−16). Convergent network strength in iNPCs was not correlated with network strength in neurons. d, Venn diagrams of the total number of unique node genes within convergent networks for each cell type. The lack of overlapping node genes between cell types (d), as well as the weak correlations of convergence strength between immature and mature cell types (c), suggest greater cell-type specificity in the magnitude of network-level convergence compared with gene-level convergence. e, One-sided enrichment ratios from ORA of cell-type-specific (color of bars) convergent node genes for rare-variant targets. FDR-based multiple testing correction was performed: #unadjusted P ≤ 0.05, *FDR ≤ 0.05, **FDR < 0.01, ***FDR < 0.001. f,g, Representative cell-type-specific network plots for convergence across 15 genes (ARID1B, ASH1L, ASXL3, BCL11A, KDM5B, CHD2, MBD5, MED13L, NRXN1, PHF12, PHF21A, SETD5, SIN3A, SKI, SMARRC2) from iNPCs (f) and mature iGLUTs (g). Network genes were filtered for protein-coding genes, clustered and annotated based on the primary node gene for each cluster. GSEA of the networks identified unique functions by cell type. Convergent networks in iNPCs were enriched for pathways associated with neurogenesis (for example, cell cycle, cell division, EPO signaling), and in mature iGLUTs for pathways associated with synaptic function (transmembrane transport and receptor signaling, secretory vesicles, SNARE complex). Illustrations in a created in BioRender; Townsley, K. https://biorender.com/efkzzf6 (2026).
