Dear followers*,
Here's a post to get to know you better!
Glad to have you around! 😃
*yes, you can still fill the poll if you're not following 😉
⤵️ Giving the final touches to the #HybridDrive
In more details:
Adding the cone
(painted with conductive epoxy, connected to the drive’s ground, to (try and) shield it from electrical interference, but it’s the first time I do this so I don’t know if it will work. Once on the rat, it will have a cap as well, except when plugged into the recording system.)
#GoldPlating
This drive has 2 sets of tetrodes: 13 microns Nichrome wire and 17 microns Platinum-Iridium wire (testing which is best). Due to different wire material and diameter, they get gold-plated slightly differently; see this for a good plating tutorial for 12-13 um Nichrome, generously shared by John Bladon: https://github.com/elduvelle/ephys_tutorials/blob/main/4_gold_plating_Nichrome_12um.md
View of the finished drive.
I am also testing the effects of implanting the tetrodes directly in the brain during surgery, or keeping them retracted in the drive and only lower them at the end of the surgery. This is for a bilateral dorsal #Hippocampus implant so it has two groups of tetrodes. I personally do not understand the point of not implanting directly in brain…
The guide/outer cannulas and their #Tetrodes
Each guide cannula holds 4 tetrodes which move together, excepted 1 cannula that holds a single tetrode - to be left in the corpus callosum and used for reference. If everything works as planned. I used 27 gauge (thin walled) inner cannulas to guide the tetrodes - the smaller diameter that you can see); these do not move, only the shuttles and the tetrodes move.
Next step is implantation surgery!
⤵️ We have signals! I think… I’ve never implanted that high in cortex. It’s definitely very silent for now. Does it look normal to you?
The timescale is 2s; this shows the unfiltered and unreferenced signals. There are a few disconnected channels.
After trying it out, I still don’t understand the point of implanting your #Tetrodes out of the brain and wait until the end of the surgery to lower them. I much prefer watching them go in the brain, so I know what’s happening! And it’s also much faster.
⤵️ It looks like we have #Theta already! Probably volume conduction - cortex above hippocampus doesn’t generate theta, right?
(This shows half of the signals)
Also, thanks to my great PI, we solved a problem of silent/attenuated signals due to a bad drive -> headstage adaptor :)
#EphysTip 1: if something feels fishy, it’s probably because it is!
#EphysTip 2: always test all parts of your circuit with a signal generator
Weird question: do you know if there's any actual publication pointing out that place cells aren't all perfect gaussian fields? I'm realizing that this might just be a completely public "secret", in that anyone who looks at hippocampal data knows it and everyone who doesn't.... doesn't
@dlevenstein yes, very good point! Goes together with another common misconception about #PlaceCells that they have to have 1 unique place field - no! This is only the case in very small environnements.
Regarding non-Gaussian place fields:
The human hippocampus has been implicated in memory, in particular episodic or declarative memory. In rats, hippocampal lesions cause selective spatial deficits, and hippocampal complex spike cells (place cells) exhibit spatially localized firing, suggesting a role in spatial memory, although broade …
@dlevenstein (and @kevinbolding) Also just for experiencing the variety of place field shapes people can check the supplementary data of our “4-room” experiment biorxiv version:
https://www.biorxiv.org/content/biorxiv/early/2020/10/21/2020.10.20.346130/DC2/embed/media-2.pdf? (That link will download the pdf, this one is the page linking to sup data: https://www.biorxiv.org/content/10.1101/2020.10.20.346130v1.supplementary-material)
@elduvelle @kevinbolding I’m curious: about how many would you say you see like the second from the left in the bottom row? 🤔
https://twitter.com/dlevenstein/status/1644722955929223169?s=46&t=zLACKZjahUD8huiRa33NWg
Here’s a nicer example with more simultaneously- recorded #PlaceCells in a relatively large rectangular arena.
Each small inset is the rate map of a place cell, ranked by spatial information I think.
Simultaneously recorded is really cool because it shows that you can get such a wide diversity of fields with exactly the same behaviour!
recorded by Roddy Grieves and shared with his approval
@kevinbolding @dlevenstein @elduvelle
Fields are definitely not Gaussian. I don't know of any place cell data papers that say they are Gaussian. Some of the early models showed Gaussian fields, but that was because they were concentrating on the issue of how to get place sensitivity at all without it being simple cue-responses. (Remember that was the first big issue with place fields - they weren't trivial sensory cells!)
For one thing, they stretch backwards along the direction of travel. For another, they can slide against walls. Here are some early place fields from our lab in the classic "open field" (which is neither open nor a field 🙄).
[Edit: these are simultaneously recorded.]
@elduvelle @kevinbolding @dlevenstein
Don't smooth your display figures! (Ever!)
Unless you are specifically making a smoothness model, you should always display your data directly. For 2D plots, that Gaussian filter can really mislead you. (For example, a single pixel noise burst looks like a beautiful Gaussian after smoothing...)
Modeling hasn't really been based on Gaussians for a while, since it is now (since the 1990s) based on combinations of internal signals (spiking relative to other fields) and external signals (which often have Gaussian noise). But Gaussian noise in the external associated cues does not a Gaussian place field make.
@adredish @elduvelle @kevinbolding @dlevenstein
(Sorry slight rant) I think this statement about smoothing is overly restrictive. It's all about choices you make for a bias-variance tradeoff and how close that gets you to the true model. If you have strong priors about how smooth the data should be (given that people have studied place fields for a long time, I think we do have pretty good priors), then smoothing makes a lot of sense because you are reducing variance (although of course if you have infinite data or a large amount of data it doesn't matter too much). There are also ways to choose the amount of smoothing via the data. Any binning is a choice of smoothing anyways.
I think the more nuanced take is make informed choices about the data and don't just do something without thinking about it. Look at the underlying data and check that your choices make sense.
Summary: smoothing != bad
El Duvelle
Dan Levenstein
kevinbolding
Redish Lab
Eric Denovellis
Chongxi Lai
Chris Rodgers