Ham Claustrum

An addendum to this is that Shane added around the brain regions, that we would need principled inter regional analysis. In this, 1) Can we predict behaviour and 2) Can we be behaviour independent, how does one region “read” another.
Slack messages
Just an FYI - Theoretical Neuroscience by Dayan and Abbott Chapter 9 has information on predicting rewards in classical conditioning tasks. Just in case we were thinking of doing anything like this. 9.3 onwards particularly could be relevant for us. Warning, there is a bit of maths around haha.

From https://app.slack.com/client/TKMCAPLAJ/GL33JCPC5/thread/GL33JCPC5-1563920927.003500

Shane O'Mara 20:23
This is virtually impossible for in vivo recording experiments. I think all you can do is have a good rule of thumb: you collect at least n cells in n animals (where is n is a decent number) - say 10-40 cells per animal in at least 4-8 animals. And you repeat the experiments as often as is possible, preferably independently. In the case of CA1, for example, you will always find about 70% of your cells are place cells. BUT! if you do a manipulation with these cells - say a remapping experiment - you need to repeat the experiment multiple times with multiple animals with multiple cells to show that the effect you are claiming is indeed there. The big advantage of doing recording experiments is that you have a base case which is founded on a survey methodology. Same is true for head direction cells - put electrodes in dorsal ATN and you will find head direction cells (say 50% or so of the cells you record). Make a claim though: for example, that head direction cells are unaffected by ambient lighting conditions - then you need to record them in the light and dark and show there is no difference in their activity between these two conditions. Good rule of thumb here: record 30-40 HD cells per animal (say 6-8 animals) and test each cell in light and dark. And then ask - are they affected by changing light conditions? It'll be obvious from the recordings, and the stats merely confirm this.

Shane O'Mara 20:35
In your case: my suggestion is straightforward. First, do the behavioural training - and let's see how easily 4-6 animals can learn the concurrent schedules. Once we know how easily they learn the task, we then do the in vivo recordings. We specify all the variables of potential interest - like I did on your whiteboard - and we decide which variables are of interest, given our thinking about what the CLA does. We set out an analysis pathway which is clear about HARKing - I think this is a bigger problem that power or p-hacking, tbh. We then do the specified analysis pathway, and test what we think is likely to be the case. And we do exploratory analyses - and then replicate the experiment to ensure that we haven't HARKed our way to a result (iow, a replication sample as they do in genomics). How many animals? The behavioural expt: 4-6; successfully implanted and targeted CLA animals - say 8-10 or so, with a hoped for yield of 10 or cells per animal. No one else in the world is doing these expts: so if we get them done, and it turns out CLA has the role we think it does, the others will want to replicate what we've done. So we must get it right! Hence we need to timeline the expt, and decide ahead of time what our hypotheses are (these can be wrong by the way!).

From https://app.slack.com/client/TKMCAPLAJ/GL33JCPC5/thread/GL33JCPC5-1562186966.005600

Synchrony
Ham has mentioned multiple times about synchrony - if I get it right I should definitely put it here.