X-Message-Number: 26838 From: Date: Thu, 18 Aug 2005 05:06:57 EDT Subject: Uploading technology (1.ii.1) The axon simulation. Uploading technology (1.ii.1) The axon simulation. A true biological neuron is far more complex that the old McCulloch and Pitts model from the 1943 era. It seems no electronics full model has been defined and built up to now. Such a neuron would be broken into a set of information processing compartments: -The presynaptic one. -The synaptic cleft. -The postsynaptic one. -The dentrite spine and dendrite segment. -The dendrite tree(s) and cell soma, including the axon first segment, the hillock. -The axon. The post synaptic domain seems the hardest to model, it will be so worked out as the last item. The axon is the simplest, so I start with it. The axon is the cell output, is is simply a transmission cable with, near the end, many branching elements. each ending in a presynaptic domain. For modern animal, birds, mammals,... the axon is myelinated that is, it is covered with an insulating membrane with only small, regularly spaced, nude domains, the so called Ranvier nodes where sodium ion channel gate are concentrated. The conduction is produced in a jumping way between these nodes. This is called saltatory conduction and correspond, on the mathematical ground to the full solution of a differential equation using the "cha" function expansion of a Green's function. The electric discharge produced by the neuron is called an action potential, its time variation is somewhat complex, but a simple rectangular pulse produce nearly the same effect at the presynaptic terminal. So, the action potential shape convey no information it seems, it is simply a by-product of the particular saltatory conduction used. In the electronics system, the neuron send out its address in place of the action potential. That address will be searched in a looking table to find all the destination synapses. A new link may be established by adding a destination synapse in the looking table, in the same way, another synapse may be suppressed by deleting it from the looking table. Given that time share processing don't simulate all neurons at the same time, there is as many destination blocks as there are time slots. For example neuron 219 broadcast its address: (addr219) in the looking table that correspond to synapses : syn12-421, syn 4023-17536, syn8-12,... The synapse 1 of neuron 421 will be simulated 3 steps after neuron 219, so the list for that step will include only syn12-421. The following two are simulated at the same time, 18 steps after neuron 219, so the 19th list will include: syn4023-17536 and syn8-12, and so on. If there is more than one chip, one list set must be keept for each one. Now, the axons are not so simple. There is one information in the duration of the action potential. Adding 20% more time may scalle up the neuromediator output at presynaptic terminal by a factor of two. Even more: This main neuromediator may be storred in small or large vesicle, the small ones contain only the main neuromediator, on the other hand, the large ones are filled with a mix of neuromediator and one or more secondary peptide mediator modulating the action of the first. The large vesicles are released only if there is a long action potential. So it is important to convey the signal of the action potential duration. This can be done by adding two or three bits to the neuron address, these will tell about the intended action potential duration. Its not all, the discharge frequency contains too an information. We have to think about action potential trains. Depending on the geometry of the axonal tree, some branching can or not convey a given frequency. For example a low frequency activity could be distributed in all the axonal tree, when the frequency goes up, some branch could not conduct it. If there are well separated impulse trains, some branchs could transmit only the first pulse of each train. So, all destination synapses of a given neuron don't see the same activity and the same message. All of that must be taken into account at the looking table level. Before it, a set of frequency filter must route the address towards different sub-looking table, each with a particular filter. The good new is that all that complexitie is fixed. Only a geometrical alteration of the axonal tree would change it. Such axonal reworking would need some protein processing in the axon, this would need a ribosomal aparatus, fortunately, it seems there is a strong barriere to the moving of ribosomes in the axon of adult brains. This is the general case, some brain area may not conform to it yet... So, even the filter set and the sub-looking tables must be programmed as variables. Yvan Bozzonetti. Content-Type: text/html; charset="ISO-8859-1" [ AUTOMATICALLY SKIPPING HTML ENCODING! ] Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=26838