neurogenesis-sim

Grow spiking neural networks the way nervous systems are built — by hand, by genes, or both — and simulate them cheaply.

neurogenesis-sim is a general toolbox for building circuits through development. You describe where tissue is (2D or 3D shapes) and what is already known about it (layers, areas, maps, cell positions). You then choose a sequence of developmental stages (cell division, migration, differentiation, axon growth, pruning, activity-dependent refinement …). For every parameter of every stage you decide whether it is set by hand, read from a hand-drawn field, controlled by genes (and then evolved), or fitted to data. The result is a spiking network that can be simulated, measured, evolved and shared as a module.

A sensor sheet growing a topographic projection onto a map

A sensor sheet → map circuit grown from a genome (see Topographic projection grown from genes).

Workflow

What you specify

What development does

Supervised (by hand)

geometry, cells, connections or connection rules

compiles and simulates exactly what you specified

Unsupervised (genetic)

a genome and a schedule of stages

genes control every stage; evolution searches genomes that work for every developmental seed

Semi-supervised

a hand-made scaffold (layers, maps, landmarks) plus a schedule

genes and fitted parameters work on top of the scaffold

For AI assistants

The whole documentation is also published as one Markdown file, with a generated status section (what exists, what is not done yet, open pull requests), to paste or upload into a chat with an AI assistant: neurogenesis-sim-context.md (about 70k tokens), or the short neurogenesis-sim-brief.md (about 11k tokens). Both are rebuilt with the site; python scripts/build_context.py makes them locally.

Highlights

  • One integer spiking kernel, four implementations with identical spikes: Python reference, JAX (CPU / GPU), the browser (TypeScript) and a portable C library (nglif) used by the desktop apps.

  • Switchable neuron models on the same grown network: integer LIF, two-compartment, Hodgkin–Huxley.

  • Know the cost before you run: memory and time per model, calibrated on your machine (python -m neurogenesis estimate).

  • Development as data: body plans, genomes, schedules of stages and parameter sources are JSON files.

  • Evolution that generalises across development: island model with fresh developmental seeds each generation and audits on held-out seeds.

  • Modules: share a working circuit (plan + genome + interface + validation) and compose it with others.

  • Viewers and apps: a 3D browser viewer, a native macOS app and a Windows / Linux app (in progress).

Project status (November 2026)

The spiking kernels, neuron models, cost estimates, body plans with the fixed growth procedure, genomes, evolution, modules, the browser viewer and the bridge are tested and usable. Configurable developmental stages, fields, geometries, fitting, connectome import, pharmacology, arbors and SONATA export work end to end and are tested; the stages are young and their API may still change (see neurogenesis-sim roadmap).

Getting started

Tutorials

Use cases

Integration

Reference

Background

Project