OpenScope Predictive Coding Analysis Map: draft stage1 for an (interactive?) navigator for multi-session neural data analysis

[Sarruedi]

I drafted an analysis map as a visual companion to the analysis outlines described in the preprint. The map organises the planned analyses across the experimental paradigms focusing on mismatch responses, prediction analysis, and across sessions Analysis, hopefully making it easier to see how individual analyses relate to each other and to the broader scientific questions.

The current version is a deliberate compromise: detailed enough to give a meaningful overview of the analysis pipeline, but not so granular that it becomes difficult to navigate. Ideally, each analysis node would include a brief description of the scientific rationale, the single-neuron and population-level approaches, key statistical tests, and links to related analyses across paradigms, along with labels for team members working on a given problem and a link to the relevant notebook or codebase.

I think there is potential to develop something like this into a fully interactive HTML guide that walks the reader through the analysis steps and can be expanded upon, essentially a living document that grows alongside the project and analysis. The current files are made in illustrator and could serve as a visual template for generating an interactive version in the next stage.

Any feedback on improvements, refinements, or expansions would be greatly appreciated!

Overview
OpenScope_AnalysisMap.pdf

A first attempt to make it a webpage with fictional placeholders for code and participating team members:

OpenScope_AnalysisFlowchart.html

PDFs
Individual Analyses
OpenScope_MismatchAnalysis_Map_SR.pdf
OpenScope_PredictionAnalysis_Map_SR.pdf
OpenScope_AcrossSessionsPredictionAnalysis_Outline_SR.pdf
OpenScope_SanityCheck_TuningProperties_Map_SR.pdf

[maierav]

This is awesome. Exactly what we need at this point. Thank you!

Some might wonder which collection of analyses could be bundled into standalone publications. To me, the color coding comes pretty close to answering that question. Though there is a lot of flexibility, of course.

[jeromelecoq]

1. I would probably agree there. Should we also list what analysis that would need to go into a data release publication, like step 0? I am assuming this would be a lot of basic thing that are required for all questions. Is that right?

• validation of cell extraction
• validation of stimuli
• exemplary notebook for each session
• ...

2. Is there a natural order of those 3 colored blocks?

3. I think Stefan also mentioned modeling as a downstream effort that should perhaps be added to the schematic?

[lrudelt]

This is great, thank you Sarah!

Your idea to make this interactive would be great to move forward and track the available resources. I was wondering, though, if HTML would be sufficiently flexible as well as plug & play for everyone.

I was thinking of Google sheets as an alternative solution. Here, it would be easy to add also the basic analysis, preprocessing and quality controls and link those analyses to the analysis tasks in your schema. Those links would make it easy to track contributions, and to see which basic analyses are missing to move forward with the other tasks. Also, this makes it easy to use spatial arrangement for structuring the work packages.

See this link for a first attempt to implement this. Note that here, solid grey boxes are things that are already implemented, and light gray boxes are basic analyses that are still missing.

This is just a first rough draft where I only added our analysis and its dependencies to get an idea. This is not comprehensive and all the other analyses are still missing.

One last point: Jerome was suggesting that we should soon aim for a publication that introduces the data plus some basic analyses and quality controls. This analysis map could be a good place to add all the missing analyses for that paper as light gray boxes within the two circles in the middle. Just an idea.

Let me know what you think!

[farznaj]

Thanks Sarah. Well done as a starting point.

Notes:

1. I am not quite clear about the different forms of novelty (multiplicative, additive, subtractive), specially the multiplicative and additive ones, and particularly about how analyses can tell them apart.

2. Analysis related to control blocks seem to be missing. [sorry if it's there and I've missed it].
   Related to this point though, I've been always unclear about the exact purpose of having all the 4 control blocks for all session types, and all recording modalities.

If it's for cross-referencing neural activities across different session types, this is inherent to 2p data by imaging the same FOV, so we don't need the same control blocks repeated across all sessions; just the one block that is relevant to that given session type.

If it is specific to NPX data, to say eg neuron 1—which responded in a specific way in a given session type—should be the same as neuron 10 in the closed loop session, because both these neurons have similar responses in the 4 control blocks?

In this case, I see why the 4 control blocks are helpful to the NPX data, but still my question remains for the 2p data and why we need them.

[Sarruedi]

Hi Lukas,

I think the Google Spreadsheet in combination with a visual display is a great idea. The rationale for a html page was just to make it interactive so that one gradually can increase the level of detail. I really like the idea of a visual analysis map per se and discussed this with Jerome. Maybe we can find a way to keep is organized as complexity is added to it.

[lrudelt]

Hi Sarah,
I am sorry, I meant to write Google Slides. With the link above, essentially anyone can edit this and add their analysis as a box, adding links to their code and corresponding slides if the results have been presented in a meeting. In principle, one could also do this with HTML, but it might be more complex to implement.

The idea would be that when a person or team adds an analysis, they draw links it all resources that the analysis has been building on, as well as the tasks that it corresponds to.

[Sarruedi]

I actually think we could combine your map with a spreadsheet to track what questions people are working on, what stage one is at and who is contributing etc. once we have that we can combine it with the Google slides visual map that otherwise will become very cluttered. If we have both we can even integrate them into an interactive page in a programmatic way. What do you think?

[lrudelt]

I really like this idea!

[Sarruedi]

Ok, then let's make a spreadsheet that we can populate and expand upon. I think this could also be useful for crafting the narrative for the data release.

@jeromelecoq are we discussing this in more detail in the next meeting?

[jeromelecoq]

Hi @Sarruedi sorry, I missed this question. Perhaps the ask is too short notice at this point. Would you be both available to discuss this on March 17?

[jeromelecoq]

Would part of the orange block belongs in a step 0 data paper ? https://docs.google.com/presentation/d/1AcIeGZM_OYsmshYT6lfQXNcyg2PRDKwzgfDtbkuSJ1o/edit?usp=sharing

RF characterization, basic tuning plots, ... Essentially the analysis of the control blocks.
What do you all think?

[lrudelt]

I have drafted a spreadsheet where we could list all the analysis packages with contributors listed, a short description and relevant links: https://docs.google.com/spreadsheets/d/1QjYpEhyf6c0zeP7t1C0Ij5IZM7dgLoouh8CUaL7qzU4/edit?usp=sharing