Work at the Pachitariu lab during summer 2020 (aka annus coronalis).
An exercise in two-photon neural data analysis and proper scientific computing practices.
We analyzed 20,000-40,000 simulatenously imaged neurons from the mouse visual cortex. Two-photon images were deconvolved into spikes with suite2p. More details here.
We inferred and characterized each neuron's receptive field (RF) using ridge regression and fitted a Gabor filter to each RF with gradient descent.
We identified linear communication subspaces using regularized canonical correlation analysis (CCA) and explored the sensitivity of the results to the number of stimuli.
We strive to use the current best practices in scientific computing and software engineering including but not limited to:
- Testing with CI/CD
- Extensive documentation
- Static type checks
- Code autoformatting
- Modular architecture
- Standardized data storage (HDF5) with parameters coupled to data
- Perceptually uniform colormaps for visualization