Flow-ID

Project Description

• Aligning calcium recordings across individuals remains challenging to neuron or cell type level. To tackle, this project will use the dandisets on C. elegans to develop approaches that can potentially be extended to more species.

Objectives and Approach

• Reproduce Sprague et al. (2025) results that 1) CRF_ID > StatAtlas and 2) more datasets included for producing the common atlas help with generalization
• Propose new alignment and ID algorithms and evaluate on benchmark against CRF_ID and StatAtlas

Progress

• Proposed and benchmarked 8 new algorithms variants against CRF_ID and StatAtlas, including normalizing flow based methods and inclusion of functional information.
• Demonstrated normalizing flow based variants win across the board, including the previous SOTA CRF_ID.

Next Steps

• Improve on functional embedding which is currently not helpful in improving alignment scores.