Thread by @eglerean

Starting now "ABC Seminar: Human Neocortical Neurosolver" by prof. Stephanie R. Jones aalto.fi/en/events/abc-… @abc_aalto

The challenge for SRJ is to connect macroscale human EEG/MEG signals to cellular and circuit level electrophysiology. "To bridge the gap is to use computational modelling"

The Human Neurocortical Neurosolver bridges methods like source modelling with invasive electrophysiology.

"Any model that simulates (brain) dynamics is a model with differential equations"

"Models can have increasing computational complexity: at the minimum we have Neural Mass Models (used e.g. in DCM)"

"We can then go to neuronal level and model spiking neurons, so called point neural modelling"

"As we know that M/EEG come primarly from the cortex, we can model the neuronal populations of the cortex and then to detailed anatomy and physiology" (see next image)

"1. When we do inverse solution we estimate the primary electrical currents (J_p)"

"2. The primary currents are generated by the postsynaptic intracellular current flow in long and spatially aligned cortical pyramidal neuron dendrites"

"with features from the cortical circuitry we can inform our model

1. Layered structure with excitatory and inhibitory interactions
2. Layer specific exogenous synaptic drive
3. Primary current dipole (from the inverse solution) comes from intracellular current flow in pyram.neu

See Fig1 here ncbi.nlm.nih.gov/pmc/articles/P…

"None of us can get anything meaningful from outside as we cannot get the dense fibres inside the cortex"

"We simulate the electrical activaty of the neurons and huge the Hodgkin-Huxley Dynamics model considering Gabaergic and Glutamaergic Synapses"

What is read from the model is the "Primary Current Dipole: Net Intracellular Current Flow in PN dendrites" (in nanoAmperemeters nAm)

see pnas.org/content/pnas/e…
pnas.org/content/113/33…

"The unique thing of this model is that it simulates the human current source signals from dendritic current flow" See the tool at hnn.borwn.edu (and it will be integrated with @mne_python !!!)

SRJ "What signals can we study with these models? It depends on what we can do with M/EEG data: ERPs, Inter-areal functional connectivity and more complex machine learning and graph theoretical measures"

"Right now we stick to the minimum level of complexity with HNN (i.e ERPs)"

"HNN reveals beta band mechanism" see jneurosci.org/content/30/41/…

SRJ: "Pre stimulus beta power is a good predictor of inhibited perception" "In general beta seems to be inhibitory for processing and we have been looking at interactions between brain areas" see jneurosci.org/content/35/5/2…

SRJ "The real question is why does beta inhibit processing?" "HNN model helps us to address this question"

SRJ "What does beta look like in UNAVERAGED data" (The brain does not average!)

"What features contribute to functionally relevant differences in average power?"

"What are the circuit mechanisms?"

"How do these mechanisms translate to inhibited function"

"Averages do weird things!" (I couldn't agree more!!) See: sciencedirect.com/science/articl…

"This is a robust phenomena with spontaneous beta oscillations at different time points:

see cdn.elifesciences.org/articles/29086…

There's even a toolbox to do the same checks of beta spectrum single trial plots github.com/jonescompneuro…

"having a peak that lasts for 50ms is way different than saying that there is an oscillation at 1/50ms (beta)" (so true!!)

"A burst signal that happens for 40ms manifests as beta band power, but it is not a beta oscillation" see figure 1 and 7 here pnas.org/content/113/33…

"There are multiple differences that gets merged (and messed) when averaging over beta" "Correlation with average prestim. beta power happens more often" See figure 3 and 6 here cdn.elifesciences.org/articles/29086…

But the question is "is it functionally relevant?" ... YES! (of course) :) See figure 7 at cdn.elifesciences.org/articles/29086…

"increasing the attention decreased the probability of inhibiting beta event happening"

"non detected trials tend to have more pre stim beta events"

(and that was shown in humans and also in mice!) elifesciences.org/articles/29086

"HNN model has many parameters, the first thing to limit the parameter space is to model the spatial physiology and then the local network connectivity between the different cortical layers"

"The only thing we tuned is the timing and strength of the driving parameters" (see next image)

"Beta events come from two inputs: a broad proximal excitatory synaptic drive and a strong distal drive that lasts a beta period (~50ms)"

"we can now have 1 to 1 mapping with features from the signals and what we get from the HNN model"

"How to test the model predictions? Invasive recordings!"

"The shape of the waveform from LFP signals were predicted by HNN in mice and monkeys"

See fig 8 here pnas.org/content/113/33…

(she is too fast even for me, but she was talking about Fig9 from the same PNAS 2016 paper pnas.org/content/pnas/1…

"Beta events emerge from a broad proximal excitatory synaptic drive" "BEta decreases M70 peak during tactile perception" jneurosci.org/content/27/40/…

"Same model is now being used for other sensory responses like auditory or visual"

She is now discussing some results in preparation about the HNN model predicting overall decrease of activity of large pyram neurons in M1.

"beta decreases the salience of sensory input via recruitment of supragranular inhibition"

Now SRJ is going to show the actual HNN tool which is available here: hnn.brown.edu it can run on multiple platforms and there are many tutorials

Next: TMS! "using TMS rather than mimicking regular oscillations, to mimic bursts of beta activity"

"Endogenous beta events share commonalities with the standard cTBS protocol" (continuous theta bursts stimulation)

And it’s over! Thank you @drstephjones!! That was A LOT of information.