NATURE | NEWS FEATURE Computer modelling: Brain in a...

M. Mitchell Waldrop

NATURE | NEWS FEATURE

Computer modelling: Brain in a boxHenry Markram wants €1 billion to model the entire human brain. Sceptics don't think he should get it.

22 February 2012

It wasn't quite the lynching that Henry Markram had expected. But the barrage of sceptical comments from his

fellow neuroscientists — “It's crap,” said one — definitely made the day feel like a tribunal.

Officially, the Swiss Academy of Sciences meeting in Bern on 20 January was an overview of large-scale

computer modelling in neuroscience. Unofficially, it was neuroscientists' first real chance to get answers about

Markram's controversial proposal for the Human Brain Project (HBP) — an effort to build a supercomputer

simulation that integrates everything known about the human brain, from the structures of ion channels in

neural cell membranes up to mechanisms behind conscious decision-making.

Markram, a South-African-born brain electrophysiologist who joined the Swiss Federal Institute of Technology

in Lausanne (EPFL) a decade ago, may soon see his ambition fulfilled. The project is one of six finalists vying

to win €1 billion (US$1.3 billion) as one of the European Union's two new decade-long Flagship initiatives.

D. BALIBOUSE/REUTERS

Computer modelling: Brain in a box : Nature News & Comment http://www.nature.com/news/computer-modelling-brain-in-a...

1 of 12 02/18/2013 11:50 AM

Related stories

Turing at 100: Legacy of

a universal mind

European researchers

chase billion-euro

technology prize

Publisher seeks patent

More related stories

“Brain researchers are generating 60,000 papers per year,” said Markram as

he explained the concept in Bern. “They're all beautiful, fantastic studies —

but all focused on their one little corner: this molecule, this brain region, this

function, this map.” The HBP would integrate these discoveries, he said, and

create models to explore how neural circuits are organized, and how they

give rise to behaviour and cognition — among the deepest mysteries in

neuroscience. Ultimately, said Markram, the HBP would even help

researchers to grapple with disorders such as Alzheimer's disease. “If we

don't have an integrated view, we won't understand these diseases,” he

declared.

As the response at the meeting made clear, however, there is deep unease about Markram's vision. Many

neuroscientists think it is ill-conceived, not least because Markram's idiosyncratic approach to brain simulation

strikes them as grotesquely cumbersome and over-detailed. They see the HBP as overhyped, thanks to

breathless media reports about what it will accomplish. And they're not at all sure that they can trust Markram

to run a project that is truly open to other ideas.

“We need variance in neuroscience,” declared Rodney Douglas, co-director of the

Institute for Neuroinformatics (INI), a joint initiative of the University of Zurich and the

Swiss Federal Institute of Technology in Zurich (ETH Zurich). Given how little is known

about the brain, he said, “we need as many different people expressing as many different

ideas as possible” — a diversity that would be threatened if so much scarce neuroscience

research money were to be diverted into a single endeavour.

Markram was undeterred. Right now, he argued, neuroscientists have no plan for

achieving a comprehensive understanding of the brain. “So this is the plan,” he said.

“Build unifying models.”

Markram's big idea

Markram has been on a quest for unity since at least 1980, when he began undergraduate studies at the

University of Cape Town in South Africa. He abandoned his first field of study, psychiatry, when he decided

that it was mainly about putting people into diagnostic pigeonholes and medicating them accordingly. “This

was never going to tell us how the brain worked,” he recalled in Bern.

His search for a new direction led Markram to the laboratory of Douglas, then a young neuroscientist at Cape

Town. Markram was enthralled. “I said, 'That's it! For the rest of my life, I'm going to dig into the brain and

understand how it works, down to the smallest detail we can possibly find.'”

That enthusiasm carried Markram to a PhD at the Weizmann Institute of Science in Rehovot, Israel; to

postdoctoral stints at the US National Institutes of Health in Bethesda, Maryland, and at the Max Planck

nature.com/turing


2 of 12 02/18/2013 11:50 AM

Institute for Medical Research in Heidelberg, Germany; and, in 1995, to a faculty position at Weizmann. He

earned a formidable reputation as an experimenter, notably demonstrating spike-timing-dependent plasticity

— in which the strength of neural connections changes according to when impulses arrive and leave (H.

Markram et al. Science 275, 213–215; 1997).

By the mid-1990s, individual discoveries were leaving him dissatisfied. “I realized I could be doing this for the

next 25, 30 years of my career, and it was still not going to help me understand how the brain works,” he said.

To do better, he reasoned, neuroscientists would have to pool their discoveries systematically. Every

experiment at least tacitly involves a model, whether it is the molecular structure of an ion channel or the

dynamics of a cortical circuit. With computers, Markram realized, you could encode all of those models

explicitly and get them to work together. That would help researchers to find the gaps and contradictions in

their knowledge and identify the experiments needed to resolve them.

Markram's insight wasn't original: scientists have been devising mathematical

models of neural activity since the early twentieth century, and using

computers for the task since the 1950s (see page 462). But his ambition was

vast. Instead of modelling each neuron as, say, a point-like node in a larger

neural network, he proposed to model them in all their multi-branching detail

— down to their myriad ion channels (see 'Building a brain'). And instead of

modelling just the neural circuits involved in, say, the sense of smell, he

wanted to model everything, “from the genetic level, the molecular level, the

neurons and synapses, how microcircuits are formed, macrocircuits,

mesocircuits, brain areas — until we get to understand how to link these

levels, all the way up to behaviour and cognition”.

The computer power required to run such a grand unified theory of the brain would be roughly an exaflop, or

1018 operations per second — hopeless in the 1990s. But Markram was undaunted: available computer power

doubles roughly every 18 months, which meant that exascale computers could be available by the 2020s (see

'Far to go'). And in the meantime, he argued, neuroscientists ought to be getting ready for them.

Markram's ambitions fit perfectly with those of Patrick Aebischer, a neuroscientist who became president of

the EPFL in 2000 and wanted to make the university a powerhouse in both computation and biomedical

research. Markram was one of his first recruits, in 2002. “Henry gave us an excuse to buy a Blue Gene,” says

Aebischer, referring to a then-new IBM supercomputer optimized for large-scale simulations. One was

installed at the EPFL in 2005, allowing Markram to launch the Blue Brain Project: his first experiment in

integrative neuroscience and, in retrospect, a prototype for the HBP.

Part of the project has been a demonstration of what a unifying model might mean, says Markram, who

started with a data set on the rat cortex that he and his students had been accumulating since the 1990s. It

SOURCE: BBP/EPFL


3 of 12 02/18/2013 11:50 AM

included results from some 20,000 experiments in many labs, he says — “data on about every cell type that

we had come across, the morphology, the reconstruction in three dimensions, the electrical properties, the

synaptic communication, where the synapses are located, the way the synapses behave, even genetic data

about what genes are expressed”.

By the end of 2005, his team had integrated all the relevant portions of this

data set into a single-neuron model. By 2008, the researchers had linked

about 10,000 such models into a simulation of a tube-shaped piece of cortex

known as a cortical column. Now, using a more advanced version of Blue

Gene, they have simulated 100 interconnected columns.

The effort has yielded some discoveries, says Markram, such as the as-yet

unpublished statistical distribution of synapses in a column. But its real

achievement has been to prove that unifying models can, as promised, serve

as repositories for data on cortical structure and function. Indeed, most of the

team's efforts have gone into creating “the huge ecosystem of infrastructure

and software” required to make Blue Brain useful to every neuroscientist, says Markram. This includes

automatic tools for turning data into simulations, and informatics tools such as http://channelpedia.net — a

user-editable website that automatically collates structural data on ion channels from publications in the

PubMed database, and currently incorporates some 180,000 abstracts.

The ultimate goal was always to integrate data across the entire brain, says Markram. The opportunity to

approach that scale finally arose in December 2009, when the European Union announced that it was

prepared to pour some €1 billion into each of two high-risk, but potentially transformational, Flagship projects.

Markram, who had been part of the 27-member advisory group that endorsed the initiative, lost no time in

organizing his own entry. And in May 2011, the HBP was named as one of six candidates that would receive

seed money and prepare a full-scale proposal, due in May 2012.

If the HBP is selected, one of the key goals will be to make it highly collaborative and Internet-accessible,

open to researchers from around the world, says Markram, adding that the project consortium already

comprises some 150 principal investigators and 70 institutions in 22 countries. “It will be lots of Einsteins

coming together to build a brain,” he says, each bringing his or her own ideas and expertise.

Bottom to top

The description of the HBP as an open user facility sparked interest and enthusiasm at the Bern meeting. But

much more vocal were Markram's critics, many of whom focused on the perceived inadequacies of the Blue

Brain model — and of Markram's approach to data integration.

At the heart of that approach is Markram's conviction that a good unifying model has to assimilate data from

the bottom up. In his view, modellers should start at the most basic level — he focuses on ion channels

SOURCE: BBP/EPFL


4 of 12 02/18/2013 11:50 AM

“It will be lots of Einsteins

coming together to build a

brain.”

Simulating the brain — the next decisive

because they determine when a neuron fires — and get everything working at one level before proceeding to

the next. This requires a lot of educated guesses, but Markram argues that the admittedly huge gaps in

knowledge about the brain can be filled with data as experiments are published — the Blue Brain model is

updated once a week. The alternative approach, approximating and abstracting away the biological detail,

leaves no way to be sure that the model's behaviour has anything to do with how the brain works, said

Markram.

This is where other computational neuroscientists gnash their teeth. Most

of them are already using simple models of individual neurons to explore

high-level functions such as pattern recognition. Markram's bottom-up

approach risks missing the wood for the trees, many of them argued in

Bern: the model could be so detailed that it is no easier to understand

than the real brain. And that is if Markram can build it at all. Judging by what Blue Brain has accomplished in

the past six years, critics said, that seems unlikely. The tiny swathe of simulated rat cortex has no inputs from

sensory organs or outputs to other parts of the brain, and produces almost no interesting behaviour, pointed

out Kevan Martin, co-director of the INI, in an e-mail. It is “certainly not the case” that Markram has simulated

the column as it works in a whole animal, he said.

Markram's response to such criticisms in Bern was that more capabilities are always being added to the Blue

Brain simulation. But Martin remained unimpressed. “I cannot imagine how this level of detail, which is still

very incomplete even after Henry's considerable labours, is ever going to be obtained from more than a few

regions of the rodent brain in the next decade, let alone brains of Drosophila, zebrafish, songbird, mouse or

monkey,” his e-mail continued.

“Of course,” Martin added, “all this would be but a storm in the professors' teacups” if the HBP hadn't come

along and raised the stakes enormously. It is all too easy to imagine other areas of neuroscience research

being starved for resources by the HBP — especially in Switzerland, which as host country will have to

provide a substantial, but still-undetermined, fraction of the funding. Douglas asks: should Europe be spending

€1 billion to support the passionate quest of one man? He concedes that visionaries are sometimes necessary

to drive progress. “But what if they're passionately wrong?”

Also fuelling anxiety — and irritation — is the widespread sense that Markram has been making his case

through the news media, not through publishing, conferences and the other conventional channels of science.

Reporters see much to like: Markram is tall, striking and explains his ideas with the clarity, quotability and

urgency of a South African version of the late Carl Sagan. He has “a hypnotic effect”, says Richard Hahnloser,

a computational neuroscientist at the INI. But critics say that this results in too many news accounts that leave

the impression that the HBP will, say, eliminate the need for experimental animals.

“The whole neuroscience community will be in trouble ten

years from now” when the implied predictions don't come


5 of 12 02/18/2013 11:50 AM

years

Henry Markram speaking at the 2011

International Supercomputing Conference

true, says another INI researcher, who worries that the

politicians will be right there saying, “But you promised!”

March of progress

In Bern, Markram bristled at accusations that he has

deliberately cultivated hype. “I have never said that the HBP would replace animal experiments,” he shot back

at one questioner. “I said that simulation helps you choose the experiments that will best add value.”

Markram was also at pains to insist that the HBP will be open to other modelling approaches. “This concern is

unfounded because they simply have not bothered to find out what is being proposed,” he told Nature after the

meeting. The final facility “will allow anyone to build models at a range of levels of biological detail with as

much data as possible from anywhere”.

Markram seems to be building support. Last year, the board that oversees both the ETH and the EPFL

enthusiastically endorsed the Blue Brain Project after a rigorous review by a four-member panel that included

two outspoken sceptics of Markram's approach. The board asked the Swiss parliament to commit 75 million

Swiss francs (US$81 million) to the project for 2013–16 — more than ten times Blue Brain's current budget.

Parliament's decision is expected next month.

Markram is optimistic that the European Union will come to much the same conclusion about the HBP.

However, if the project isn't endorsed, says Markram, “we'll just continue with Blue Brain” — although it may

take a lot longer to reach a full brain simulation.

Markram clearly feels that history is on his side. “Simulation-based research is an inevitability,” he declared in

Bern. “If I get stopped from doing this, it's going to happen. It has happened already in many areas of science.

And it is going to happen in life science.”

Nature 482, 456–458 (23 February 2012) doi:10.1038/482456a

Related stories and links

From nature.com

Turing at 100: Legacy of a universal mind

22 February 2012

European researchers chase billion-euro technology prize

08 March 2011

Publisher seeks patent

07 May 2010

Bioinformatics: Industrializing neuroscience

10 January 2007

The Blue Brain Project


6 of 12 02/18/2013 11:50 AM

Matthew Kirkcaldie said:

Ian Watson said:

01 February 2006

Blue Brain boots up to mixed response

08 June 2005

Nature Special: Turing

From elsewhere

The Blue Brain Project

The Human Brain Project

The European Flagship Initiatives

Henry Markram's 2009 TED talk

Henry Markram at the 2011 International Supercomputing Conference

Comments

Every account I've seen of Blue Brain has omitted glia from consideration,

or mentioned them as an afterthought on the to-do list. What's the point of modelling neurons down to

the channel if the main regulators of the ionic environment, and the complementary functional partners

of neurons, are simply ignored or assumed?

Douglas and Martin have written some of the most penetrating insights into cortical function in print – if

they're prepared to go on the record as sceptical, there's a very good reason for it. Simulation is the

playing out of the consequences of existing data and ideas – the best it can reveal is hidden

implications, or limitations, of those facts. To suggest that simulation could replace research into actual

nervous systems is absurd: we're not trying to understand the model, we're trying to understand the

biology. If the model throws a quirk, it may be interesting, it may be misleading, or it may be wrong –

there's no way to tell without recourse to biology. If a living nervous system throws a quirk, it tells us

we are not getting it, and we need to try harder – and may open the door to some completely

unexpected phenomenon. Given that we don't have even the slightest idea of the basis of

consciousness, the most interesting and unique thing that neurons do, I highly doubt our ability to

identify the deepest essentials of neuron function for modelling.

"It is the tension between creativity and skepticism that has produced the stunning

and unexpected findings of science." - Carl Sagan

The HBP seems very similar to the Grand Engineering Challenge identified in 2008 by the US National

Academy of Engineering to reverse-engineer the human brain and the $21 DARPA funded project

SyNAPSE So the real issue is should the Europeans join these endeavours. However, it has to be

said that there are questionable scientists, like Ray Kurzweil involved with these projects.

2012-02-22 07:48 AM

2012-02-22 10:53 AM


7 of 12 02/18/2013 11:50 AM

Basil Weibel said:

Mitchell M. Waldrop said:

Paul Adams said:

Arnold Trehub said:

I'm publishing a book about Alan Turing called The Universal Machine and write more on these issues

in my blog

I don't know anything about the field. However, do I see it correctly that all the

critics mentioned in the article are affiliated with the INI Zurich which is clearly an institution competing

for the same funding money? I miss the voice of an totally independent expert.

Basil,

To set the record straight: it is true that the only critics mentioned in the article are from INI. But that is

because they were the only ones willing to talk on the record. All the other critics (and supporters) that

I interviewed — and there were many — asked not to be quoted by name, precisely because the

debate has gotten so sensitive. So instead, I tried to indicate the range of opinion in the community by

my description of the comments at the Bern meeting. It was an imperfect solution, to be sure, but the

best I could do in the circumstances.

Mitch

The key issue is whether the somewhat stereotyped circuitry of the neocortex

implements hitherto rather elusive biological principles, built on but going beyond standard information

theory. Markram does not claim to have an inkling, but clearly hopes that it might be easier to discover

such principles in an inevitably approximate, oversimplified and inaccurate cortical model than in the

real thing. As critics note, it might be more productive to pursue instead a variety of competing ideas,

however far-fetched: without principles and new ideas, as Rutherford said, the whole enterprise

degenerates into stamp-collecting. He also said â€œWe've got no money, so we've got to thinkâ€.

If Markram can convincingly explain how a computer model can simulate a

human brain representing a volumetric space from a fixed locus of perspectival origin — the neuronal

foundation of our phenomenal world — then the project might be justified. Otherwise not.

2012-02-23 05:12 AM

2012-02-23 07:49 AM

2012-02-24 07:27 AM

2012-02-27 09:05 AM

2012-02-28 11:03 AM


8 of 12 02/18/2013 11:50 AM

Martin Richardson said:

Andy Prybyszewski said:

Andy Prybyszewski said:

This is a good idea just from the perspective of providing a framework.

Experiments tend to compare certain qualities partly because of the statistical tools used in analysis of

results. A global model permits the creation of smaller models as objects, and their integration into an

overall framework.

The brain is such a large network that it is expected that such a model will reveal some interesting

possibilities. Research has been bogged down in allocating function to specific areas of the brain,

even though it is well understood that location and function are flexible and can be changed in the

event of injury which would indicate a more holistic nature of consciousness.

This is a great idea and I strongly support Henry's project to put a great

amount of knowledge related to neuroscience together. However, I have problem with his approach to

reasoning.

I give a simple example from my experiments in the retina, which anatomy and physiology is well

known and which is isolated CNS structure from the top down influences. We've got unique

intracellular recordings from cat retinal GC "in vivo" (Vis.Res.1993, 33, 861-875). I spent about half of

year to simulate the spike generation process that converts intracellular generator potential into spike

train. The problem is that even if we know ion channel dynamics in the axon hillock all data (e.g from

salamander, rat,etc) are recorded in the temp about 20C or lower. We can adapt ion channel dynamics

to 36C by Q10 factor but it is not clear that this is universal constant for each ion channel and what is

the variability of ion channels density in different neurons. I have succeed to simulate EVERY single

spike timing in output of the retina (Biol.Cyb, 1996, 74: 299-308) and on this basis an analog (not

Digital) model of the retina was built (IEEE Tran NN, 2007, 18:70-85).

As Henry noticed in his Science (1997, 275, 213-215) paper every spike can by the coincident

mechanism change a synaptic efficacy in the cortex. In order to simulate this mechanism and to

compare with experimental data we need such data, but as I know nobody cares about single spikes.

We do not have such recordings from every cell in the cortex.

Generally it is a problem with a huge number of parameters that must be determined experimentally. I

think that each neuron is an analog computer performing variable (as Henry noticed) spatio-temporal

computations and a single spike it the result of the neuronal computations and not a noise.

This is a great idea and I strongly support Henry's project to put a great

amount of knowledge related to neuroscience together. However, I have problem with his approach to

reasoning.

2012-03-10 11:46 AM

2012-03-10 01:00 AM


9 of 12 02/18/2013 11:50 AM

David Dos said:

Kristen Salesman said:

Evgeney Knyazhev said:

I give a simple example from my experiments in the retina, which anatomy and physiology is well

known and which is isolated CNS structure from the top down influences. We've got unique

intracellular recordings from cat retinal GC "in vivo" (Vis.Res.1993, 33, 861-875). I spent about half of

year to simulate the spike generation process that converts intracellular generator potential into spike

train. The problem is that even if we know ion channel dynamics in the axon hillock all data (e.g from

salamander, rat,etc) are recorded in the temp about 20C or lower. We can adapt ion channel dynamics

to 36C by Q10 factor but it is not clear that this is universal constant for each ion channel and what is

the variability of ion channels density in different neurons. I have succeed to simulate EVERY single

spike timing in output of the retina (Biol.Cyb, 1996, 74: 299-308) and on this basis an analog (not

Digital) model of the retina was built (IEEE Tran NN, 2007, 18:70-85).

As Henry noticed in his Science (1997, 275, 213-215) paper every spike can by the coincident

mechanism change a synaptic efficacy in the cortex. In order to simulate this mechanism and to

compare with experimental data we need such data, but as I know nobody cares about single spikes.

We do not have such recordings from every cell in the cortex.

Generally it is a problem with a huge number of parameters that must be determined experimentally. I

think that each neuron is an analog computer performing variable (as Henry noticed) spatio-temporal

computations and a single spike it the result of the neuronal computations and not a noise.

If I would have this billion I would give it! But imho it is impossible. Just read what

wikipedia writes about the Human brain

Is a billion-euro model of a brain worth the money???

The type of casement doors that would ensure one of the most stylish designs to your home is the

exterior french door

For now, most problem isn't to find tonnes of money to run project, but

there always has been grave question of technologies itself & applied methods, i.e. how those

methods are looked through prism of moral laws. Just let's remember the prime obstacle of

mathematical modeling always has been to synchronize multithreading beast safely. + It's not top

secret that reliable code takes performance rate drastically down & let's not forget man-hours to keep

all-that going on. At now, we need project to experiment on human deeply to get most clear data. I see

it sounds too awful... Just to stop righteous question, answer: I'm ready to volunteer 4 that.

2012-05-23 01:47 AM

2012-08-01 10:58 AM

2012-08-24 06:52 AM


10 of 12 02/18/2013 11:50 AM

Nature ISSN 0028-0836 EISSN 1476-4687

Dennis Hollenberg said:

Lester Ingber said:

The problems with this effort appear both fundamental and staggering.

Firstly, they fail to realize that the architecture they assume appropriate to their simulation will not

allow tests using potential architectures.

Secondly, these researchers should ask themselves the basic question: what architecture likely

organizes the brain (hint: what 'architecture' existed before organisms?).

Thirdly, simulation requires a high level of abstraction in order to produce indications of profitable

research avenues (assuming they get #2 right); in modelling at the synapse level they appear to be

aiming at too fine a target and stand to reap an overwhelming tsunami of data that they will be unable

to process and organize. Here's a recent quote in which Markram's describes the desired shade of

blue sky:

"If their current bid for â‚¬1 billion ($1.3 billion) of European Commission funding over the next 10

years is successful, Markram predicts that his computer neuroscientists are a decade away from

producing a synthetic mind that could, in theory, talk and interact in the same way humans do." CNN

Oct. 12, 2012.

Lastly, I'm afraid that, without any demonstrated theoretical model guiding it, this is merely a

make-work or employment program.

The scale of the BB simulation is both too coarse and too fine. It ignores the

importance of multiple scales of interactions in the real brain. For example, the importance of

large-scale coherent firings among many macrocolumns, as they influence sensitive balances among

ionic mechanics, which drive synaptic and astrocyte interactions, in turn influencing the large-scale

coherent activity, is ignored, thereby skipping both larger and smaller scales by BB.

You need to be registered with Nature and agree to our Community Guidelines to leave a comment. Please

log in or register as a new user. You will be re-directed back to this page.

2012-10-12 08:04 AM

2013-01-23 09:51 AM

See other News & Comment articles from Nature


11 of 12 02/18/2013 11:50 AM

NATURE | NEWS FEATURE Computer modelling: Brain in a...

Documents

Transcript of NATURE | NEWS FEATURE Computer modelling: Brain in a...