Pages

March 05, 2008

Blue Brain status and the future of whole brain simulation

Seed magazine has a 9 page review of the Blue Brain project. Blue Brain is the IBM project to simulate a human brain on a supercomputer. Currently they have simulated one column of a neocortex (of a rat) with 10,000 neurons and 30 million synapses using a 22.8 teraflop supercomputer. (A human neocortex column has 60,000 neurons.) Their simulation uses 400 segments for each neuron and they have precisely researched individual ion channels and biological functions to precisely generate the simulation. Their simulation is generating its emergent results from ground up matching to physical measurements.

I think this project has similarities to the human genome project. There is value and things to be learned in having a precise functioning computer model of a human brain and even parts of a human brain. There will be even more to learn when we can affordably make many such models for different people. A follow on goal of the human genome project is to sequence the genomes of 100,000 people, which is a step towards to goal of everyone having a copy of their personal genome. The follow on to Blue Brain or other projects to simulate the human brain is to simulate the brains of different types of people and then personal brain simulations. In parallel would be whole body simulations down to the intracellular and then molecular level. After you have working systems you can reduce the complexity where it is unnecessary. For example 1% of the genome is where important differences between people is located. Of course long running and molecularly accurate personal brain simulations would be equivalent to mind uploading.

A whole human brain has 100 billion neurons and 1 trillion synapses. So they need to scale up the neurons by 10 million times and the synapses by 33,000 times. They believe the computing power to do this will be available in less than 10 years. So by 2017, there should a single whole brain simulation. Personalized whole brain simulation would follow by 2027-2037.

Benefits of the blue brain project

Gathering and Testing 100 Years of Data

Cracking the Neural Code

The Neural Code refers to how the brain builds objects using electrical patterns. In the same way that the neuron is the elementary cell for computing in the brain, the NCC is the elementary network for computing in the neocortex. Creating an accurate replica of the NCC which faithfully reproduces the emergent electrical dynamics of the real microcircuit, is an absolute requirement to revealing how the neocortex processes, stores and retrieves information.


Understanding Neocortical Information Processing

A Novel Tool for Drug Discovery for Brain Disorders

A Global Facility [to test theories of brain function]

A Foundation for Whole Brain Simulations

A Foundation for Molecular Modeling of Brain Function

Now that the column is finished, the project is pursuing two separate goals:
- construction of a simulation on the molecular level, which is desirable since it allows to study effects of gene expression;
- simplification of the column simulation to allow for parallel simulation of large numbers of connected columns, with the ultimate goal of simulating a whole neocortex (which in humans consists of about 1 million cortical columns).

FURTHER READING
UPDATE: reader svante pointed out the need for non-destructive brain scanning for personalized brain simulation. It would be needed and below is a survey of the current status of that work.

Neuroimaging now

There has been precise monitoring of animal brain function via 2 photon microscopy. They used a clear window into living brains. If the functional end of the microscopes could be a lot smaller and placed under the skull then that would gather a lot more of info needed for personalizing the scans.

There is also detection of chemical reactions in living human cells

Other advances in faster and more sensitive cell viewing

MIT group working on making brain cell activity monitoring

An optical microscope has better than 10nm resolution

FURTHER READING
Al fin comments about this bottom up approach and the progress on top down analysis of brain function and the merging both efforts.

6 comments:

svante said...

There will be some difficulty in "personalizing" this technology, as the analysis is done by making thin slices of the brain, which pretty much will kill you if you are not already dead. That is a major difference from genome analysis, I would say. Granted, some new technology might emerge to enable in vitro brain scanning, but that is highly speculative.

bw said...

Correct that non-destructive scanning would be needed.

Neuroimaging now

There has been precise monitoring of animal brain function via 2 photon microscopy. They used a clear window into living brains. If the functional end of the microscopes could be a lot smaller and placed under the skull then that would gather a lot more of info needed for personalizing the scans.

There is also detection of chemical reactions in living human cells

Other advances in faster and more sensitive cell viewing

MIT group working on making brain cell activity monitoring

An optical microscope has better than 10nm resolution

Anonymous said...

I read the article (a bit quickly) but unless I missed it, it doesn't answer an important question: Who is financing this project? Anyone know?

bw said...

I believe funding is mostly from IBM and the Swiss government.

EPFL gets funding from a variety of sources including Nestle. Your chocolate habit, helping to fund the Singularity.

Nestle gives 25 million swiss francs to the brain and mind institute

Anonymous said...

brain has at least 100 trilions sinapses. They can't personalize worm or mouse, but suspect to personalize human...
I read that Bluebrain project is IBM advertisement.

qraal said...

Hi Brian

Do you think the brain simulation will need to be so detailed? The information needed to emulate the brain has a lot of wildly varying estimates and I often wonder how fine-grain an emulation needs to be.

Sure, there's roughly 30 billion neocortical neurones in an adult male, and about 25 billion in a female, and they can have up to 10,000 synaptic connections to other neurones, but we know so little about how information is actually represented in the brain - could be a bit per synaptic connection or per neurone.

Then there's all the neuroglial cells wrapping around each neurone, all with their own slow-paced signalling systems - I wonder how they're being emulated?

Definitely a fascinating project to watch unfold. I think in vivo real-time brain-cell monitoring is perhaps not too far off if the MRI technology using ambient magnetic fields can be adapted to the skull. Any word in that direction?