Discussion: meetings/2025-08-26

[jeromelecoq]

I am preparing figures to support our discussion

Here is a summary figure for the first production sessions on neuropixels and mesoscope.
https://docs.google.com/presentation/d/1LlD1HZpzUtsEExsRZ-TLYgoz16xNBGdAVlLI_5Hsqvw/edit?usp=sharing

[jeromelecoq]

alternatively to the RF mapping, we could use this very new stimulus. I am curious what people think
about that ; https://www.biorxiv.org/content/10.1101/2025.07.19.665666v1.full.pdf

[Sarruedi]

Sounds potentially very useful indeed.

[jeromelecoq]

Sounds potentially very useful indeed.

Should we invite some of the authors to present? Those are some of your UCL colleagues, right Sarah?
I looked a little bit and I don't think there are available generated videos, yet, right?

[jeromelecoq]

My presentation is here : https://docs.google.com/presentation/d/1LlD1HZpzUtsEExsRZ-TLYgoz16xNBGdAVlLI_5Hsqvw/edit?usp=sharing
as indicated above

[maierav]

my slides are here: https://docs.google.com/presentation/d/1TIxrYdJ_JltmIK2Nq79d84stuwdrn3Z68G3EFn3lhjY/edit?usp=sharing

[jeromelecoq]

@maierav I am following up on our discussion.
I am not sure I answered correctly to your question during our meeting today. The code intent is to construct the list of trials to be presented and THEN shuffle them. We are not pulling parameters randomly at each presentation. This is pseudo-randomisation as you described it, correct?
see here :

openscope-community-predictive-processing/code/stimulus-control/src/Mindscope/generate_experiment_csv.py

Line 612 in b2b1695

random.shuffle(trials)

This is exactly why we are creating the stim. table prior to the session: To guarantee a fixed number of presentations trials, while ensuring randomization of presentation order. Are you concerned about other statistical effects?

[maierav]

Yes, exactly. That's perfect!

https://docs.python.org/3/library/random.html

[jeromelecoq]

Hi all,
We are doing adjustment to the targeting plans for the neuropixels recordings.
We found that Switch RL to VISa is compatible with LGN recording. Essentially one of our probes can go exactly through the middle of both.
We think this is a superior design based on feedbacks and are planning to do this.

[jeromelecoq]

Severine will share screenshot of these.

[Sarruedi]

Thanks for sharing both configurations, Séverine. For the pilot, if the priority is reliable coverage of the middle of dLGN, I agree the VISa configuration is the safer choice. It gives us better margin for centring in dLGN, whereas the RL trajectory is quite anterior to LGd and risks skimming the edge of LGN. Functionally, VISrl is earlier in the dorsal hierarchy than VISa and closer to early motion and optic-flow processing, so it would be good to discuss whether switching to VISa has any implications for the specific hypotheses we want to test. Conceptually, it would favor RL given its position along the hierarchy and properties. If we need the geometry to center in LGd for the pilot, VISa is good choice. We just need to be aware that the aictvity may reflect more integrated prediction signals

[koweiss]

@severine2305 @jeromelecoq
I think the Vis a configuration is better for thalamocortical investigation and has higher certainty to target dLGN.
@Sarruedi I agree about the hierarchy, but I'm not sure if the differences in integrative properties would be that different. Sensorimotor tasks in particular engage many areas simultaneously, and recurrent connectivity makes both RL and VISa worth investigating. However, I think it's valuable to select a probe trajectory that targets a mostly feedforward connectivity (dLGN) and a recurrent/feedback one (VISa), at least to contrast bottom-up and top-down signals in predictive coding.

[maierav]

I might not be qualified to comment given my minuscule knowledge about murine cortex, but I lean with @koweiss.

Based on our prior work, earlier areas seem to signal "prediction erros" mostly as "release from adaptation" (i.e., local sequence deviations only - even if they are highly predictable).

We found contextual, global deviant detection only at the top of the hierarchy (matching earlier work: https://www.pnas.org/doi/full/10.1073/pnas.1117807108).

So, sampling mid- to higher-tier areas might be beneficial to investigate PP that goes beyond basic sensory adaptation.

There is also some debate, suggesting that LM, RL etc. really are just one and the same area (V2) - though I have only followed that debate peripherally, and I know this is contentious.

Lastly, getting a good shoot at the LGN sounds like it may provide us a valuable "control" condition:

If we find PP in V1, the question arises whether PP originates there, or is inherited from the LGN (or even the retina).

So, being able to analyze LGN responses would be of great help in that case. This having said, I am not familiar with the retinotopy of mouse LGN and thus which angle provides the highest chance of covering the visual space we stimulate...

[Sarruedi]

Thanks for the feedback @koweiss and @maierav. I agree having dLGN is potentially very valuable. So in that sense VisA is an excellent compromise for a higher visual area to record from

[severine2305]

Hi everyone, so we will go with the vis a/LGd implant. Thanks for your feedback!