Sam McKenzie

Au revoir Twitter. Should have left long ago. To my scientist friends and colleagues, looking forward to meeting you IRL.

@kareem_carr We had it in DC public school. Latin too.

Americans need to start teaching formal logic in the schools. I'm 100% serious about this. It's a national emergency.

@DrewBHeadley At first we obsessed over latencies, but the long window needed for state classification relaxed that. we use the sockets provided by intan rhx to pipe data into matlab. they offer nice control for distributing jobs for async processing. Our biggest latency is the USB protocol

@samamckenzie Sounds fantastic! Curious what hardware you are using for the real-time detection.

Tomorrow is a big day for the lab. Often these threads mark the end of a project, but I'd like to think this is more the end of the beginning. 1/N

And one more thought. If you have a seizure forecasting algorithm and want to compare it against ours. Get in touch. we can try it once we are done.

Being able to do these kinds of experiments was a big reason I came to @UNMHSC , so thank you to the CBRR and @BillShuttleNM . Hopefully, I will be able to follow up with a proper manuscript. Until then, wish us luck!

We don't know if the DBS protocol will reduce seizures, but for the first time we have a chance of starting to know. Of course there will be more controls and troubleshooting - the end of the beginning. 8/N

finally, we have a subject for whom we have trained a robust prediction model; I observed it working in real-time today! We optimized the DBS parameters over the last couple of weeks. Tomorrow, for the 1st time, we (my tech Davey Gregg and myself) are ready to close the loop 7/N

Next we had to get the rat epilepsy model right. Enough seizures to train our models and statistically detect treatment effects, but not so many that prediction is trivial. 6/N

Next, we had to find a brain target where stimulation could induce a therapeutic effect within our forecasting horizon and ideally one in which the protection outlasts stimulation. we did this, and developed a Bayesian optimization algorithm to get the stim parameters right. 5/N

First we had to develop an algorithm. We did this on a large retrospective dataset generously provided by @Chris_Lisgaras and @scharfmanlab . then we had to get it running in real time. 4/N

There are several DBS treatment options for those with drug-resistant epilepsy. Some are responsive and deliver current after seizure detection. Wouldn't it be great if DBS started prior to seizure onset. Treatment where and when you need it to stop seizures before they begin 2/N

The seizure forecasting field has had this dream for 10+ years, but better ML algorithms and datasets have made a difference. we feel it is time to start developing DBS protocols that are effective given prior to a predicted seizure (these likely differ from those we know) 3/N

“I hear you are buying an RFID system and data loggers and throwing out your Master-8 computer because you want to make something real. You want to make a Panksepp paper.”

@yutasenzai So cool! Congratulations Yuta

Finally, to establish causality between SC activity and virtual heading, we blocked activity in the right SC that contributes to left (CCW) turns in awake mice. Strikingly, this abolished CCW shifts in virtual heading during REM sleep, leading to biased virtual head turns. 5/7

New preprint post! We show that motor commands in the superior colliculus shift the internal representation of heading during REM sleep despite the immobility of sleeping mice. Thus, the brain simulates actions and their consequences during REM sleep.🧵1/7 doi.org/10.1101/2024.0…

@UeliRutishauser @Nature Would be interesting to try your analyses on our rodent dataset pubmed.ncbi.nlm.nih.gov/24910078/

Delighted our latest finding! We discovered that abstract representations emerge in the human hippocampus when learning to perform inference. This change in neural geometry is due to disentanglement of discovered latent and observable variables. @Nature nature.com/articles/s4158…

@AndrewHires @patrickmineault 178.8 as a r/g color blind

@patrickmineault 172.5 Boundary

@dlevenstein pubmed.ncbi.nlm.nih.gov/34650039/

@dlevenstein Try changing the I/O gain. Let me know if you have predictions for how place fields should differ. They get sharper with experience, but maybe they are always sharp and the neuromod milieu distorts the attractor. Maybe only attractor is replayed thus poor replay of early moments

Very pleased to share our first experiments looking at how norepinephrine controls the segmentation of episodic memory biorxiv.org/content/10.110…

Much more work to do-does CA1 control its own NE release? if so through terminal control of polysynaptic drive to LC? Does NE move hippocampal codes away from an attractor? is the correlation we observed between NE and representational uniqueness causal? stay tuned

It doesn't in fact, we found that there was enhancement of replay for moments further away from context transition. We hypothesize that the NE effects on neural coding creates a unique neural timestamp of the moments after transition that may contribute to the primacy effect