Canon PowerShot SD700 IS
ƒ/2.8
5.8 mm
1/60

Following up on last year’s IBM Almaden Institute, I went to visit Paul Rhodes at Evolved Machines.

This whiteboard with RC ladders and diff eq’s looked strangely familiar (tickling the EE in this geek).

To more accurately model neuronal signaling, the dendrite is broken up into a finite element model of adjacent segments. Somewhat like a coax cable, the dendrite has an inner core of conductive ions surrounded by a lipid layer that insulates the signal from the common ground of the intercellular fluid. In addition to static membrane capacitance and axial resistance, the myriad ion channel pores penetrating the lipid insulator provide rich non-linear dynamics (modeled as a battery and voltage-regulated resistor). Conformal proteins in the channel dynamically regulate sodium and potassium current, with positive feedback accelerating the local rise of voltage above a threshold, and then reversing to rapid negative feedback (creating a classic neuronal voltage spike).

The voltage spike will propagate, with gain, along the dendrite, from one finite element region to the neighbor, not as a free flow of electrons, but as a bucket brigade of opening and closing ion channels, like fingers cascading down a very long flute.

The dendrite with myriad branches is a fundamental locus of computation, not just the neuron cell body.

I am looking forward to Cognitive Computing 2007 at Berkeley.

10 responses to “Mental Model”

  1. That conference looks like a lot of fun… make sure to have a good chat with Lotfi Zadeh. I’ve always wanted to meet the guy, he did most of the original work in fuzzy logic. Certainly a brilliant and unconventional thinker!

    I’ve been going to a number of neural computation conferences, and I am consistently underwhelmed by the lack of intermediate scale simulations. There are plenty of people doing classical computation of single and several neurons, and some people are doing hand-wavey ‘just-so’ simulations of very large scale computation, but it is hard to find anyone doing decent work of intermediate circuit scales. A shame because most of the really interesting wet-computation (IMHO) takes place across tens of thousands of neurons.

    This is in part because of the difficulty in obtaining decent information about both structure and function at those scales. Single cell electrodes work well, as do EEG, but there is a biiiiiiig gap between those two. Techniques like voltage sensitive dyes are encouraging, but they are notoriously touchy and involve lots of cutting which neurons don’t like.

    Please post recordings! Especially if you find any of the talks particularly insightful. Super super especially if anyone discusses biologically realistic simulations of decent sized numbers of neurons!

  2. will do.

    Does your work help fill the gap? Combining active models of single neurons with destructive network scans of cortical columns? Do you get ion channel detail in your scans?

    Henry Markram from EPFL used an IBM BlueGene supercomputer at 22TFLOPS to model 10 million dynamic synapses for 10K neurons…. but it still seemed very crude.

    Sister72: oh oh…you have entered a recursive loop. Look away before you Snow Crash

  3. Question: Is an electrical engineer geek an EEG?

    "The dendrite with myriad branches is a fundamental locus of computation, not just the neuron cell body."

    Yeeeeeeeesssssss! That´s it. I think! I can´t explain why, you do better at that, but it is true. I have a glimpse: Intelligence resides in the logistics of information, and the more the connections and channels (and so their quality in-site), the better the logistics. Technology is about creating, transforming but mostly *transporting* information (energy in its diverse manifestations) in incresingly more efficient ways. The brain is outstandingly technologically efficient and advanced from this point of view (mine) -even with its failures-. From a nanotechnological point of view too, no?

    * I wonder if above where I say "intelligence resides in the logistics of information" we would replace "intelligence" for another word, more abarcative… "evolution" perhaps? "Evolution resides in the increasing efficiency of the logistics of information"

  4. Yes, our work is aiming at that middle section, to have anatomically accurate mappings of large numbers of columns.

    Right now we aren’t getting ion channel detail… but we have a design in the pipeline that is pretty amazing. It will allow us to decon/reconstruct a cortical column, and then go in and grab the contents of a synapse or cell body and look at the transcriptome or proteome expression. This could be used to populate the parameters in a large model like Markram is starting to do.

    If you want to run some of your own simulations, it ain’t that hard. (Well, it ain’t hard to do it wrong in the right direction, but doing it ‘right’ persists in eluding our efforts). A nice one to use is GENESIS… http://www.genesis-sim.org/GENESIS/ And some good tutorials… http://www.genesis-sim.org/GENESIS/Tutorials/index.html

  5. Thanks for posting the GENESIS stuff Oddwick.

  6. d’oh! i think i just whacked my brain against an asymptote!

  7. How are you supposed to reproduce the Ranvier Nodes of the neural axonal body?

  8. Hey, I used this shot in my blog post here:
    laconicreply.wordpress.com/2008/09/03/i-love-finding-new-…

    In keeping with creative commons I linked to this page, and your profile for accreditation.

    Let me know if that’s cool, or if you’d like it removed.

    Thanks
    Eric

  9. the dynamic simulation is quite pretty in the end…

    Brainstorm

Leave a Reply

Your email address will not be published. Required fields are marked *