What will it take to achieve the simulation of the human brain?


20 Jun 2011 Hamburg - First keynote speaker in the opening session at ISC'11 was Dr. Henry Markram, Director of the Brain Mind Institute at EPFL in Lausanne, Switzerland. Dr. Markram has an experience of more than 20 years in brain research. He started out in medicine where he catalogued diseases and specialized in neurology in Cape Town, South Africa. There are 500 clinically classified diseases in the brain. Unfortunately, the pharmaceutical industry are pulling out of brain research because it is too costly and this industry is only focusing on where the money is. Academia receives 5% of the available funding to address the rest of the brain diseases. At present, there are some 5 million papers studying the brain which involves an enormous data growth. Dr. Markram emphasized that we need to start getting organized. The solution to the data tsunami is integration: the creation of a unifying model.

Dr. Markram stressed the importance of simulation to look for patterns and rules. Finding rules speeds up the insight and provides very accurate representations. Modelling becomes easy if you follow biology. There are 58 morphological classes. As an initial strategy, scientists need to computationally clone the neurons. To achieve this, you have to grow the neurons. The method consists in synthesizing the neuronal morphology and build the external morphology but also the internal structure.

You can define the internal organs of a single cell by using genetic information. The interaction of proteins within the cell is important. There are some 100 types of reaction to classify but if you go further into detail, the parameters collapse, explained Dr. Markram. Scientists would need a supercomputer to simulate the brain at the molecular level. Perhaps, this might not be possible but the supercomputer can model the individual cell.

Cells produce an electrical pattern. Some have a delay and some have bursting fires. We can define 16 types of character. If the scientists goes into an individual cell to study the interaction of proteins, he can detect and build different types. This amounts to ion channel combinations, distributions and kinetics. This is done by automatic modelling to study the model behaviour.

There is a distributor for the positioning of cells. Some 100 billion neurons exist and we can observe a distribution of all neurons in the whole brain. Dr. Markram thinks that constraint programming to position the neurons in the brain can mean a solution. When you start adding maps, the model collapses.

In Lausanne the team is experimenting with reverse engineering of the Connectome. As such the pattern of the synapses can be discovered. You have to look at how the fibers are attaching. There are 100 million structural touches. 98% fitted the experimental data. Scientists are selecting fibers to form synapses. To create a synapse they apply the neuron-neuron connections.

The connectivity is derived from the rules of biology. The morphological diversity ensures a robust and invariant Connectome. As such artificial neural networks and neuromorphic processors are being created. You can push a button and export the moel of the circuit. Once you have the neurons and the connection, the next step is to create a dynamic synaptic transmission.

Dr. Markram described the synapses as filters to capture the communication. At present, we have the equations that describe the transmission so now the scientists are preparing for the Blue Gene simulations.

Another aspect is the visualization using the Neuromesh generator with the Blue Hub to generate an integrated framework.

The validation consists in spontaneous slow oscillations in the cortical column. These are stimulus treshold dependent oscillations. The simulation shows local electrical field potentials. The challenge is to build a causal chain of events leading to cognition. The question is whether the model can support any intelligent action such as sensing, learning, memory, behaviour etc.

Dr. Markram explained that as a next step the team starts combining columns to build a brain region. There are many different ways to build the human brain. This has led to a brain simulation facility roadmap. In 2005 a single cellular model saw the light and in 2023 a cellular human brain model is planned. Therefore, global integration is required.

There are multiple representations and formats. For this capability the scientist has to be taken in the loop. A prototype facility is needed to build the neural circuitry. To this aim visual steering and interactive supercomputing are necessary tools.

Dr. Markram next expanded on the Human Brain Project which has received a funding of 1 billion euro over 10 years time. With this funding two simulation cockpits have been created: a HPC cockpit and a simulation cockpit. One of them is situated in Lausanne. This generic system is open to the community as a Simulation as a service: a tool to study the brain.

The Blue Brain Project is chaired by Bob Bishop and managed by Felix Schuermann and Sean Hill.

Dr. Markram is planning the so-called Dr. Neuron initiative to integrate volunteer computing in the study of the human brain. In this vision, science and donated computer power will go hand in hand to unveil the working of the human brain.
Leslie Versweyveld