Three.js · Connectomics · Visualization · 2025

Convis: Drosophila Neuron Visualization

A web tool for exploring the FlyWire fruit-fly connectome: thousands of 3D neuron morphologies linked to a classification Sankey and a UMAP of all 139,000 neurons.

Type: Scientific Visualization · Connectomics
Data: FlyWire FAFB v783 · 139,273 neurons
Stack: Three.js · ECharts · Plotly · Vite · Python
Code: github.com/lengyuner/ConVis
Date: 2025 · ongoing

Overview

Convis is an interactive web tool for exploring the structure of the fruit-fly (Drosophila) brain for neuroscience teams. It loads the FlyWire FAFB v783 connectome, with 139,273 classified neurons, and draws their skeletons in 3D next to two overview panels that stay in sync with it: a Sankey diagram of the flow, super-class and class hierarchy, and a UMAP embedding of connectivity or morphology. Selecting neurons in any one view highlights them in the other two.

Rendering at scale

The opening scene shows the first 100 neurons of each class, 2,221 in total. Skeletons are simplified offline and quantised into one packed binary, so the browser reads each neuron on demand instead of parsing thousands of SWC files. For the web build I pruned that cache from 485 MB to 19 MB by keeping the initial scene plus a class-balanced sample of small neurons, so the site can be hosted statically. A volume mesh of the brain and a neuropil selector give the neurons anatomical context.

Exploring and explaining

Neurons can be recoloured by flow, class, sub-class, cell type, hemilineage or side, and expanded to their upstream and downstream partners. External data such as gene expression or time series can be imported and mapped onto the neurons with a colour map. A chat panel sends the current selection to a Python pipeline that builds a statistical report, then asks an LLM to interpret it.