
Galen Lynch
Scientist II, Allen Institute Accelerator for Neural Dynamics
GitHub · Google Scholar · CV
I’m a neuroscientist at the Allen Institute Accelerator for Neural Dynamics in Seattle, working with Josh Siegle and Karel Svoboda. I want to understand how neural networks produce flexible, learned behavior: how animals go beyond repeating what has worked, and use what they’ve learned about the world to generate new decisions and actions.
My current work asks how prelimbic cortex and its inputs track the value of actions as mice forage. To record from those regions together, we developed brain-wide, MRI-guided electrophysiology: we image each mouse with MRI before surgery, plan Neuropixels insertions to reach small, deep targets in that individual brain, and afterwards trace every probe in cleared, light-sheet-imaged tissue. I also write much of the software behind it, from image registration and probe planning to deep-learning track tracing and data formats for terabyte-scale recordings. Much of it is open source.
For my PhD with Michale Fee at MIT, I studied the two ends of motor learning in songbirds: the variable, exploratory song of young birds and the precise, stereotyped song of adults. I found that premotor neurons in the songbird analog of motor cortex tile song continuously in time rather than clustering around vocal gestures. To record from the circuit that generates song variability, I built a lightweight microdrive for singing birds and co-developed a head-mounted microscope. More on the songbird work →
Selected projects
- Brain-wide, MRI-guided electrophysiology
Recording prelimbic cortex and its inputs at once
Imaging each mouse with MRI before surgery cuts median targeting error from 430 to 167 µm, enough to record prelimbic cortex and six of its small, deep input regions at once.
- Action value in prelimbic cortex
Foraging behavior, and brain-wide recordings of PL and its inputs
Foraging mice decide whether to leave a spout from that spout’s reward history alone, and prelimbic cortex and several of its inputs carry that history before each choice. Ongoing work.
- Exploration and skill in songbird motor learning
PhD work with Michale Fee at MIT
Ensemble recordings in singing zebra finches, on how the brain produces the variable song of young birds and the stereotyped song of adults.
- Electrode tracer
Finding Neuropixels tracks in whole-brain light-sheet volumes with deep learning
A learned 3-D detector and decoder, cross-validated on 188 annotated brains, recover dye-labeled probe tracks from cleared whole-brain light-sheet images, turning hours of manual annotation into checking an editable first draft.
- Rutter
Constrained-optimization planning for multi-probe insertions
An interactive planner and a GPU-batched optimization solver that find collision-free multi-probe insertion plans maximizing coverage of target brain structures.
- Research with coding agents
How I run long investigations with Claude Code
An append-only, pre-registered lab notebook lets a coding agent carry a months-long ML investigation across sessions without losing the thread or overstating results.
Selected publications
- 2024
- 2023An optical design enabling lightweight and large field-of-view head-mounted microscopes Nature Methods 20(4), 546–9
- 2021Neural representations of space in the hippocampus of a food-caching bird Science 373(6552), 343–8
- 2016Rhythmic continuous-time coding in the songbird analog of vocal motor cortex Neuron 90(4), 877–92