Scroll or pinch to zoom, drag the background to pan. Click an edge or a legend row to highlight that gene pair everywhere (click empty space to clear); hover a node to highlight its edges. Click a node to add it to the selection; click a selected (dark) node or its chip to remove it. Drag nodes to tidy the layout. Each type's genes are its most type-specific cell-surface/synaptic genes by P·G (mean over 50 train solves of the system's best model, projected on all genes). Edges are the real connectome, width ∝ synapse weight; each shows its top pre-gene × post-gene pairs by P·G product. A read-out of the model's cell→type assignment, not fitted β pairs. Known binding partners only: pairs restricted to the Interactome v3 list (either partner can be presynaptic), both genes P·G ≥ 0.1 in their type, ranked by co-expression enrichment log₂(P·Gpre · P·Gpost / atlas means), as in Fig 1D; dashed grey = connected types with no expressed binding pair. Visual: the atlas stores the DIP genes under their CG identifiers; they are shown as DIP-α…λ (official symbols, NCBI Gene).