search

AllenInstitute / openscope_databook

OpenScope databook: a collaborative, versioned, data-centric collection of foundational analyses for reproducible systems neuroscience 🐁🧠🔬🖥️📈
https://alleninstitute.github.io/openscope_databook
Other
63 stars 19 forks source link

Create a tca notebook #288

Closed jeromelecoq closed 6 months ago

jeromelecoq commented 1 year ago

There is a python library associated with this paper : https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6907734/pdf/nihms-976117.pdf

https://github.com/neurostatslab/tensortools

rcpeene commented 1 year ago

Got TCA to run, but the components don't look great. Current task; subselect the data to find better results

jeromelecoq commented 1 year ago

Jerome to ask in Boston at Alex in person

jeromelecoq commented 1 year ago

From Alex: Hi Jerome,

It looks great! More interpretation of the factors you get out would be useful.

I think the reason you need a lot of components is that the natural movie stimulus evokes very rich / high-dimensional temporal dynamics. Potentially it might lead to more interpretable factors if you look at static natural images with shorter trial duration. And then you could also mix in several different image types, which different trial factors would pick up.

A long time ago (probably 7 years ago, before we even published the paper) I ran TCA on some Allen data and found that some of the trial factors correlated with the animal's running speed. It would be good to check for something like this.

Also, this is a nice method to try out if you haven't seen it already: https://www.biorxiv.org/content/10.1101/2023.03.01.530616v1

-- Alex