The use of computing during real-time data streaming and multi-purpose parallel computing allows to look into an individual cell with electrical recordings. This is a whole new field in neuroscience, stated Dr. Schürmann. The idea of setting up such a complex simulation is to find out how an elementary building block is likely to behave and to visualize the process. To this purpose, the team required massive, immediate access to computing power as well as a combinatorial contextual memory.
To achieve their goal, the researchers used reverse engineering and the so-called multi-neuron patch-clamp recording technique. Giving the fact that there are 100 million nerves in the human brain, the recording of the electrical properties amounts to a huge task. The Blue Brain team performed the work in a week. The researchers are working with combinations because you are much more likely to find what you are looking for in this way, explained Dr. Schürmann.
He showed the audience what is called butterflies of the cerebral cortex. These are shapes of nerves which appear as a host of different looking species. The team has identified the morphological classes of neocortical neurons into two different groups. As such, you have the exciting cells and a whole zoo of other cells. What is interesting now to observe is the counteraction of these two types.
Head of the research at the Swiss EPFL is Prof. Dr. Henry Markram. He led his team in mapping the electrical behaviour of neurons. The 100mV voltage spikes are action potentials. They are showing the diversity of electrical behaviours in the neocortical column and thus provide a recipe of cells in the microcircuit. They serve as minimal building blocks to build, simulate and visualize the neocortical column.
The Blue Brain Project wants to consolidate the whole process in one model. There are 10.000 morphologically complex neurons and about 1000 compartments per neuron. For the passive properties the cable-equation is used. For the active properties Hodgin-Huxley (HH) differential equations are applied with terms of up to 30 types of ion-channels per compartment, as Dr. Schürmann explained.
There are 30.000.000 dynamic synapses displaying facilitation or depression according to their activity. The outputting membrane voltage per compartment amounts to 1000*10.000*10Khz*4Byte=4000B/s. The ultimate goal is a biological real-time computing with reduced output.
To this end the team has performed a powerful connection between the BlueGene and SGI supercomputers at the institute. Another task is to build tools and models to set up the complex simulation by means of the Blue Brain Project toolchain. In fact the project has to develop all the elements of the toolchain in order to build new models. In this regard the BlueBilder is an important step requiring much computing power, according to the speaker.
The researchers are trying to find 3D co-ordinates of touch points. The problem is to find proximity between 10 to the 14th - 10 to the 16th cylindrical elements and that iteratively. Because the touch points have to match morphometric statistics, such as the type of connecting neurons, the number of touch points, the location of innervation (axon initial segment, soma etc.), the challenges are enormous. The synapses are formed on a Layer 5 Pyramidal Neuron.
Statistically the team has to learn to know how the nerves connect and therefore is visualizing synaptic maps onto neocortical neurons. The compute engine Neuron is a well-established, flexible tool in neuroscience and solves the HH equations and cable-equation for each compartment in discrete time steps.
The simulations have been succesfully ported to the BG/L supercomputer but have not been optimized yet, explained Dr. Schürmann. The NCC problem is well balanced for the massively parallel BG/L architecture and BG/L's virtual node mode is well suited with essentially 1 neuron per processor but also larger circuits can be computed.
The factor which is different from many HPC applications is that the transient dynamics are of interest. The simulation I/O is limited. The first simulations of the Blue Brain Project show 10.000 neurons, 10 million dynamic synapses, morphological and electrical types. They also provide a first visualization of activity in 10 percent of the neurons.
Dr. Schürmann told the audience that the team's ambition is to realize "in silico" imaging which involves averaging dendrites across volume elements. It would be nice to visualize the activity of 1 neuron in the neocortical column.