as long as I have the attenion of the HPC folks-- what are we doing these days when bootstrapping for place cell mutual info? shuffling positions and keeping spike times (and thus ISI) constant, or shuffling spike times? Seen both in papers and unsure if there is a consensus
@aheadofthenerve I would say shuffle the spike trains (so not randomly shuffle the times of spikes but shift them all of the same amount for each shuffle to maintain firing statistics but disrupt the relation of the firing with location). Also I would make sure to use high within-session stability as an additional criterion for place cells. Looking forward to a sneak peek of your data :)
@elduvelle_neuro yeah, rn I am circularly shifting the position data relative to spikes (to keep ISI the same-- so equivalent to what you wrote), but have also seen actually shuffled position (as opposed to shifted) as well as shuffled spike times. UG!
I'm writing up the paper now so hope to have a preprint out soon(ish). All the data etc is there but it's dragging on fiddly details like how to shuffle that won't really impact the conclusion but I want it to be as bullet proof as possible, obviously
@aheadofthenerve I don’t recall seeing position shuffling or even shifting for characterization of place cells but maybe that’s for calcium imaging when you don’t have access to spike? I would say the standard is what I said above for ephys data. Shuffling position seems bad. Shifting position seems like it should be exactly the same as shifting times no?
@elduvelle_neuro yes shifting position is same as shifting cell times. but I have def also seen shuffling position or even spike or calcium transient times
@aheadofthenerve @elduvelle_neuro
Ya I think since MI is symmetric theyre the same, but wouldnt be the same with usual NHST models that have an asymmetric outcome variable. Shuffling bins is definitely more correct than shuffling times bc it preserves the distribution as u say in OP tho.
Add comment