← All projects

Inferring brain organoid structure from electrode recordings

Reverse-infer the structure of a brain organoid from multi-electrode recordings using computational models and machine learning.

  • Theme: Neuroscience, machine learning, computational modeling
  • Repo:

Team

Why this project

Recovering structure from spiking patterns is a foundational neuroscience problem and a great test bed for hybrid model + ML approaches. Public recordings exist, and a base modeling toolkit is available on site.

What a team could build in one day

  • Load the organoid recordings.
  • Fit a baseline model — e.g. connectivity inference from spike cross-correlations — and evaluate against the structural ground truth.
  • Visualize inferred vs. true structure.

Minimum viable demo: a working pipeline from recording → inferred connectivity → comparison plot for at least one organoid.

Stretch directions

  • Deep generative models of latent neural dynamics.
  • End-to-end structure inference from raw recordings.
  • Per-organoid variability and population-level summaries.

Resources

  • Methods preprint: bioRxiv 2023.12.29.573646.
  • Recordings on DANDI. Note: the dataset is currently being republished under a new ID due to federal compliance changes — check the DANDI archive for the live URL at the time of the hackathon.

Compute

High — multi-electrode recordings are large; budget for storage and a GPU for any deep-learning extensions.