Skip to content

EXASPIM-PLAN.md

Latest commit

 

History

History
67 lines (48 loc) · 4.3 KB

EXASPIM-PLAN.md

File metadata and controls

67 lines (48 loc) · 4.3 KB
Raw

Reverse-engineering notes — Neuron Morphology Community Portal

Site: https://morphology.allenneuraldynamics.org/ Captured: 2026-06-10

TL;DR

  • Frontend: Single-page React app (webpack bundle), MobX for state, Apollo Client for GraphQL, Semantic UI (CDN) for styling, Three.js + WebGL canvas for 3D.
  • API: Single GraphQL endpoint at POST /graphql (same origin). A small GET /system returns runtime config.
  • 3D viewer: Neuroglancer (confirmed by ?neuroglancer=… query params and a console warning about the google-brainmaps credentials provider). One <canvas> element, neuron skeletons + CCF reference atlas rendered together.
  • Bulk data: Neuroglancer "precomputed" format served as static files from
    • AWS S3: s3://aind-neuron-morphology-community-portal-prod-o5171v/ngv01/ (neurons — split into full/, axon/, dendrite/ segmentations, each with info, segment_properties/info, skeleton/info).
    • Google Cloud Storage: gs://allen_neuroglancer_ccf/ccf_test1/ (Allen CCF reference atlas — info, mesh/, segment_properties/).
  • 132 published neurons total. Each neuron has a UUIDv7-style ID; routes look like /neuron/<uuid> and /candidates.

Network capture (after page load + a Search click)

App bundle:

  • GET / (HTML shell) → loads bundle.js, code-split chunks 717.bundle.js, 471.bundle.js.
  • GET https://cdnjs.cloudflare.com/ajax/libs/semantic-ui/2.4.0/semantic.min.css
  • GET https://fonts.googleapis.com/css?family=Lato:...

Application API (own origin):

  • GET /system → 
    {
  "systemVersion": "3.0.11",
  "precomputedLocation": "s3://aind-neuron-morphology-community-portal-prod-o5171v/ngv01",
  "doiHandler": "https://doi.org",
  "exportLimit": 20
}
  • POST /graphql — all metadata, search, and per-neuron data.

Bulk/3D data (cross-origin, fetched directly by Neuroglancer):

  • https://aind-neuron-morphology-community-portal-prod-o5171v.s3.amazonaws.com/ngv01/{full,axon,dendrite}/info
  • …/ngv01/{full,axon,dendrite}/segment_properties/info
  • …/ngv01/{full,axon,dendrite}/skeleton/info
  • https://www.googleapis.com/storage/v1/b/allen_neuroglancer_ccf/o/ccf_test1%2F{info,segment_properties/info,mesh/info,mesh/997:0,mesh/997:0:0}?alt=media&neuroglancer=…

GraphQL schema highlights (captured from in-flight requests)

Operations observed:

  • UserQuery → user { id authDirectoryId firstName lastName emailAddress affiliation permissions }
  • ConstantsQuery → systemSettings { apiVersion neuronCount }, neuronStructures, nodeStructures { id name swcValue }, queryOperators { id display operator }, atlasStructures { id name acronym aliases structureId depth parentStructureId structureIdPath defaultColor hasGeometry }
  • GenotypesQuery → genotypes { id name }
  • CollectionsQuery → collections { id name description reference specimenCount }
  • SpecimensQuery($input: SpecimenQueryInput) → specimens { totalCount items { id label notes referenceDate referenceDataset { url segmentationUrl } tomography { url options { range window } linearTransform { scale{x,y,z} translate{x,y,z} } } … } }

So the GraphQL service is the single source of truth for metadata, search filters, and per-specimen pointers; the actual neuron geometry and atlas meshes are pulled by Neuroglancer from precomputed cloud buckets via the URLs that GraphQL hands out (e.g. referenceDataset.url, segmentationUrl, tomography.url).

Detected client-side globals

__reactRouterVersion, __mobxInstanceCount, __mobxGlobals, __THREE__, __APOLLO_CLIENT__ → React + React Router + MobX + Three.js + Apollo confirmed.

Implications for "exaSPIM-style" planning

If we want to mimic this stack:

  1. Host neuron / volume data as Neuroglancer precomputed layers on S3 (already what AIND does).
  2. Stand up a GraphQL (or any) service that returns, per specimen: referenceDataset.url, segmentationUrl, tomography.url (+ transforms, windowing, etc.). Atlas structure tree comes from the same service.
  3. Embed Neuroglancer in a React shell; let it fetch directly from S3/GCS — no proxying needed because precomputed buckets are CORS-enabled.
  4. Use the exportLimit / precomputedLocation style /system endpoint pattern for runtime config (avoids baking S3 paths into the bundle).