Connectomics: wiring as data

A map of connections

A connectome represents reconstructed neural connectivity. Let each neuron be a node and each directed synaptic connection an edge. Several synapses between a pair may be stored separately or aggregated into a weight. Choose and document that convention before computing statistics. An adjacency matrix is a coordinate system for a graph, so a row/column mistake can reverse every pathway in an otherwise correct program.

From images to a graph

Electron microscopy supplies images from which processes are segmented, traced, proofread and annotated. Each stage introduces uncertainty. Split errors can divide one cell; merge errors can join unrelated processes. A boundary can truncate connections. Treat release versions and reconstruction scope as part of the data, not a footnote added after the analysis.

Structure is a constraint, not a complete simulator

Synapse counts do not supply membrane constants, receptor effects, conduction delays, current activity or learning history. Building a neural model from a graph therefore requires extra assumptions. A good experiment changes those assumptions systematically and asks which conclusions survive. The architecture can be biologically grounded while the dynamics remain an engineered approximation.

Neural-network perspective

Here use A[i][j] for the connection from source i to target j. The input to target j is the sum over sources. Dense arrays store all pairs, including absent edges; sparse edge lists store the connections you actually use. On a large graph, format and orientation can determine whether a simulation is feasible.

SEE THE RELATIONSHIPS

From structure to behavior.

MAKE THE IDEA PRECISE

The mathematics.

Build the equation, step by step

1. For every target j, collect contributions from all source nodes i. Multiply each source activity by its outgoing weight to that target.
2. Sum those contributions to obtain y_j. With sources stored in rows, a column-vector implementation uses the transpose of A.
3. Apply the operation twice to represent two propagation steps. This computes weighted paths of length two; it is not the same as moving signals through all edges instantaneously.

WORK THROUGH THE NUMBERS

Worked example

A synthetic chain has weight 2 from node 0 to 1 and weight 3 from node 1 to 2. Starting with activity only at node 0, one propagation step reaches node 1 and the next reaches node 2. This example intentionally has no decay or spike threshold.

A small Python experiment.

Python 3 · Standard library only · Synthetic educational example. Run this in your own notebook or a local file.

edges = [(0, 1, 2), (1, 2, 3)]
x = [1, 0, 0]
for _ in range(2):
    y = [0, 0, 0]
    for source, target, weight in edges:
        y[target] += weight*x[source]
    x = y
    print(x)

Expected output

[0, 2, 0]
[0, 0, 6]

Change one quantity, predict the result, and then run the code again. Explain any difference before adding complexity.

THE NEXT LAYER

Chapter depth map.

This introduction provides a first model. The following chapters are outlines for deeper future editions; their full treatments have not yet been published.

Representing structure

Asking graph questions

Read with the original sources.

These readings support the concepts and deeper investigation. The worked numbers and illustrations above are teaching examples.