Sources

Region hierarchy, names, colours and centroids for the mouse (1,327 structures) and human (1,839) atlases. Mouse centroids are real coordinates in the P56 reference space; the human graph has none, and its regions are laid out diagrammatically.

Projections between regions, from fluorescent tracer injections registered into a common space. Strength is a normalised volume of labelled axon — a measure of WIRE, not of synapse count or functional coupling.

Digitally reconstructed neuron morphology in SWC. The neuron level draws real traced dendrites and axons, not a procedural tree.

Experimental atomic coordinates for receptors and channels. The molecule level is an alpha-carbon trace of a real solved structure.

Predicted structures where no experimental one exists. Flagged explicitly wherever shown: its per-residue values are pLDDT confidence, which runs OPPOSITE to a crystallographic B-factor.

Protein identity, gene names, length, function summaries and cross-references to solved structures. Restricted to reviewed Swiss-Prot entries.

CONSTRUCTED from literature dimensions — a 600 nm bouton, a 20 nm cleft, a 300 nm postsynaptic density, 40 nm vesicles. The numbers are real; the arrangement is a model. There is no queryable archive of synapse geometry to draw from.

2,336 real neurons, each with 56 measured features from whole-cell recordings in slice at 34 °C. This is what the models on /validate are scored against, using the same feature definitions on a simulated trace as Allen uses on a recording. A patch recording reads the membrane potential through the electrode's series resistance, so measured spike peaks sit somewhat below the true value.

Membrane potential, gating variables, ionic currents and spike propagation are SOLVED LIVE from the 1952 model. Conduction velocity comes out at 17.8 m/s against Hodgkin and Huxley's computed 18.8, and it converges across four grid spacings and three timesteps — so it is the model's answer rather than the mesh's. An earlier version reported 12.5: the gating variables were advancing blind to the axial current, so they lagged the depolarisation arriving at each compartment. The spike had the right shape, amplitude and afterhyperpolarisation, and was simply slow.

Traub & Miles rate constants as parameterised by Pospischil et al. (2008), with their published conductance densities — not values chosen to make the comparison come out well. Squid kinetics at 34 °C are not a mammalian neuron: the Q10 correction speeds every gate twentyfold, sodium inactivates before the spike develops, and the peak collapses to 8.9 mV against a real 23–41.

A thousand cells after Brunel (2000), with Bi & Poo (1998) spike-timing-dependent plasticity. The spike's SHAPE is discarded and only its timing kept, which is the trade that makes a population this size tractable. An earlier version delivered so little current per synapse that it was a thousand independent cells driven by noise; it produced entirely plausible rates.

A CATALOGUE of neurophysiology datasets in NWB format. The recordings themselves run to terabytes and are read with a proper NWB client; nothing here streams them, and the dataset list is not connected to the simulated spike trains shown elsewhere.

Regional transcript abundance, answering a question the molecule view raises and cannot settle: whether a receptor is actually present in the region you zoomed out of. It is RNA, not protein — nTPM is normalised transcript abundance, an imperfect proxy for how much protein is there.

Molecules associated with a receptor. Searched by NAME rather than by target identifier, which is fuzzy — some results are connected to the target by a mention rather than by a measured activity, which is why they are labelled 'associated with' rather than 'ligands of'.

Connectome dataset metadata. A fly, in a mouse-brain app, on purpose: no mammal has a synapse-resolution wiring diagram at whole-brain scale, and the honest way to say so is to show what one looks like where it exists. Only metadata is read — querying actual connectivity needs an auth token this deployment does not hold.

AMPA, NMDA and GABA-A conductances as peak-normalised dual exponentials, with the Jahr & Stevens (1990) magnesium block and Tsodyks–Markram (1997) short-term plasticity.