Discussion: meetings/2025-09-23

[jeromelecoq]

Here you can add material related to the meeting, as well as discuss the meeting content.

[jeromelecoq]

I will start the meeting with a few slides for context to help guide the discussion : https://docs.google.com/presentation/d/1vJ2BA1BkXF9-ucBvkN_JSAm0XmIBrhx3eIxXCdOiRNw/edit?usp=sharing

[jeromelecoq]

Thanks Dimitri for presenting. !

[jeromelecoq]

Here is the slide I wanted to discuss :

[jeromelecoq]

Here is my rationale:

• I think Zebra has been tested and compared with classical tuning stimuli + it gives more reliable responses for short stimuli. So I value adding it.
• Regardless, having a short Gabor RF mapping block would be helpful to confirm. We know we can do this in 5min with Neuropixels data.
• Trippy gives a unique link to the EM data data (excitatory + inhibitory). So I think we should replicate the exact same movie, including the seeds. The 5min seeds were also shared across all Microns sessions, reinforcing the bridge.
• The Receptive block with Gabor is a control for this pilot. We could remove later on.
• I also think, as Alex mentioned, that we could essentially apply the Wavelet decomposition from Zebra to all stim. This is related to one question I had today. Probably we will have to truncate some frames. The 15s discontinuities (the trial reset every 15s) in the microns trippy stim. are likely to have slightly difference response structure (by the way, those are quite worth looking into separately).

[maierav]

I believe I can extend a continuous Trippy stimulus to full five minutes, without the 15 seconds reset. In that case, we will use direct replicability with MiCRONS of course.

Also, what about the Monet noise? It sounded like it is still a bit unclear what the two non-overlapping populations are, but what if they turn out to be informative? So, it sounds worth considering to run Trippy + Monet instead of Trippy + Zebra.

[jeromelecoq]

1/ But wouldn't it be better to replicate what Microns did exactly? They actually have the same monitor we do.
2/ The monet is likely triggering global motion detection pathways. It is possible those two stim somehow recruit feedforward and feedback projection pyramidals which we know are completely segregated. https://pubmed.ncbi.nlm.nih.gov/21618232/
3/ We have some limit on stim duration for the session so there is a limit to how much we can add up. We could swap the RF block with Monet for the production set if that is better?

[jeromelecoq]

@maierav, I am working on implementing the bonsai movie display now.

I think I would suggest this is what we need?

• 1 min of Zebra, that we will replay 5 times across all our mice/session (exactly the same movie).
• 2.5 min of Trippy movie with 15s trials recorded as one movie, taking the seeds from Microns. We would replay it twice across all our mice/session (exactly the same movie).
• 2.5 min of Monet movie with 15s trials recorded as one movie, taking the seeds from Microns. We would replay it twice across all our mice/session (exactly the same movie).

Probably 30Hz movie at 1920, 1080

[maierav]

Python translation of the code for Monet noise:
https://colab.research.google.com/drive/1rZSzxD8eYuR6byY_SqK9gJ4NNDCCstH5?usp=sharing

[jeromelecoq]

It looks to me like the seeds to microns movies are available to download. But the only way to do this, as far as I can tell is to actually install the entire functional nda database here : https://github.com/cajal/microns-nda-access

This requires downloading 110 GB of data (functional data included), spinning up the database, running the docker. Then once this is running, it looks like it is possible to use the provided Jupyter Notebook to access the seeds.

I would estimate, it would probably take a day to set this up. If someone already has that docker set-up, it might be much faster.

[dimitri-yatsenko]

We host a free, read-only instance of the MICRONS release database. Here are the environment variables you have to configure on your client side to access the database.

ENV DJ_HOST db.datajoint.com
ENV DJ_USER microns
ENV DJ_PASS microns2023

Here is a complete tutorial that uses CodeSpaces, so you can run it right on GitHub.

https://github.com/datajoint/microns_phase3_nda/

Please let me know if you run into any trouble.
Dimitri-

[jeromelecoq]

I think I am finding that the seeds used were like so :

rng_seed n_sessions n_trials shared
1.0 6 38 True
2.0 6 38 True
3.0 6 38 True
4.0 6 38 True
5.0 6 38 True
6.0 6 38 True
7.0 6 38 True
8.0 6 38 True
9.0 6 38 True
10.0 6 38 True
18.0 1 2 False
27.0 1 2 False
52.0 1 2 False
83.0 1 2 False
91.0 1 2 False
157.0 1 2 False
169.0 1 2 False
364.0 1 2 False
440.0 1 2 False
538.0 1 2 False
729.0 1 2 False
761.0 1 2 False
774.0 1 2 False
853.0 1 2 False
854.0 1 2 False

[jeromelecoq]

This means for Trippy seeds 1 to 10 were used for the shared movies?
Is that your recollection @dimitri-yatsenko ?

[dimitri-yatsenko]

I will check with the source data.