Share 0
Like 0
Modeling the human brain

Modeling the human brain

Henry Markram tells how his son's autism fired his ambition to unlock the secrets of consciousness by using 'big data' to trace the electronic signals that zing between neurons
In a spartan office looking across Lake Geneva to the French Alps, Henry Markram is searching for a suitably big metaphor to describe his latest project. "It's going to be the Higgs boson of the brain, a Noah's archive of the mind," he says. "No, it's like a telescope that can span all the way across the universe of the brain from the micro the macro level." 

We are talking about the Human Brain Project, Markram's audacious plan to build a working model of the human brain – from neuron to hemisphere level – and simulate it on a supercomputer within the next 10 years. When Markram first unveiled his idea at a TEDGlobal conference in Oxford four years ago, few of his peers took him seriously.

The brain was too complex, they said, and in any case there was no computer fast enough. Even last year when he presented a more detailed plan at a scientific meeting in Bern, showing how the requisite computer power would be available by 2020, many neuroscientists continued to insist it could not be done and dismissed his claims as hype.

Today, thanks to the largesse of the European Union, which awarded Markram €1bn last year to make his dream a reality, many of those naysayers are being forced to take him seriously. The gift, which comes on top of a state-of-the-art IBM Blue Gene computer from the Swiss government, makes Markram's unit at the Swiss Federal Institute of Technology in Lausanne the biggest dog on the neuro block. It also gives Markram a headstart on brain-mapping projects in Japan and the US, where Barack Obama is hoping to persuade Congress to award $3bn to a similar initiative called Brain (so far Obama has pledged $100m).

The timing of Obama's initiative and the EU's award, the largest in its history, has led to talk of an international "brain race". But Markram argues that a better parallel is the Human Genome Project. Just as the decade-long effort to map the 3.3 billion base pairs that make up the 23 chromosomes in the human genome required close co-ordination between scientists worldwide, so Markram argues mapping the human brain in all its neural complexity will take a similarly co-operative international research effort.

The problem, as he sees it, is that neuroscience has become hopelessly fragmented. Each year sees the publication of about 100,000 papers, but neuroscientists are so specialised they have trouble understanding each other. We know a lot about the organisation and interaction of individual neurons and there have been countless studies, using functional magnetic resonance imaging (fMRI), of brain regions at the scale of tens of millions of neurons, but we have little information about the scales in between. Nor do we have an integrated understanding of how events at the level of genes, proteins and synapses cascade through the brain to produce behaviour and cognition. Markram points out that, using conventional approaches, it takes 20,000 experiments to map a neural circuit. Yet, in all, the brain contains 86 billion neurons. On top of that, to fully understand the operation of every synapse and how they interact with neurons in other parts of the neo-cortex, scientists would need to trace all of the 100 trillion connections between them – something that is impossible to do experimentally.

But what if, instead of trying to map these neural structures piece by piece, we could tease out some underlying principles governing their morphology and architecture? What if we could use a supercomputer to run thousands of statistical simulations so as to predict the way that those neurons are likely to combine and then check the resulting models against real data from biology? Then in theory we could predict those structures and use them to reverse-engineer the human brain. That, in a nutshell, is the principle behind the Human Brain Project and the vision that drives Markram.

"The fact is we are never going to experimentally map the human brain and people who think otherwise are deluding themselves," he says. "Instead, we have to search for the fundamental principles and then use those principles to construct a hypothesis of the bits of the brain no human has ever seen and no human will ever see. Then we have to test those hypotheses and refine the principles until our model gets better. Otherwise, we are just stabbing in the dark."

When Markram speaks this way, it is easy to see how he raises other scientists' hackles. Markram's belief in the ability of computing technology to solve the big questions of neuroscience is messianic. It is a messianism he combines with the tousled good looks of an ageing matinée idol and an undeniable charisma that at TED in Oxford four years ago had some members of the audience spellbound.

In a field dominated by big brains and even bigger egos, each mining their own esoteric field, Markram's big data approach to experimental neuroscience represents a cultural revolution. "We're saying look, if you think you're going to understand the brain on your own forget about it. We're going to have to work very differently. We're going to have to work in teams, in swarms. To someone who is used to deciding what experiment they should do I can see how that might come across as antagonistic."

Markram's belief in the need for teamwork is rooted in his own experience as a brain researcher and his conviction that only neuroscience is capable of solving the deeper mysteries of how the electrical signals zinging between neurons produce consciousness and how interferences or malfunctions in those electrical channels produce disordered or "diseased" thinking. Markram credits the awakening of his interest in these subjects to a Latin teacher at the boarding school in Durban where, at the age of 13, he was sent from the rural South African farm where he was raised. Within a term, he had changed from a rugby playing jock with little interest in schooling to being obsessed with biology and trying to understand how a slight alteration in the biochemistry of the brain could lead to conditions such as depression and schizophrenia.

This interest in biology led him to an undergraduate degree in medicine at the University of Cape Town where he hoped to specialise in psychiatry. But he quickly became disillusioned with what he saw as the discipline's lack of scientific rigour. "Psychiatry completely shocked me. The idea that you could write a computer programme to classify mental diseases and on that basis prescribe a medicine struck me as crazy."

Instead, Markram switched to experimental neuroscience and in 1985 joined a Cape Town laboratory directed by a young researcher named Rodney Douglas. It was Douglas, now an emeritus professor at the Swiss Federal Institute of Technology in Zurich, who first fired Markram's enthusiasm for lab work and, with his exceptionally steady hands – useful when stitching together neurons smaller than a pinhead, Markram was soon enjoying a meteoric rise. A PhD at the Weizmann Institute of Science, one of the leading research universities in Israel, was followed by postdoctorates at the National Institutes of Health in Bethesda, Maryland, and the Max Planck Institute for Medical Research in Heidelberg, Germany. Then, in 1995, Markram was lured back to the Weizmann as an associate professor.

There he earned a formidable reputation as an experimenter, becoming the first researcher to patch two connected neurons simultaneously. This put him in a position to see how they interacted in response to differently timed electrical signals.

A fundamental postulate of neuroscience is Hebb's rule – that neurons that fire together wire together. However, Markram discovered that the strength of these connections varies according to when impulses arrive and leave. If an input spike of electrical current occurs before an output spike and within a certain time window, on average the input connection was strengthened. However, if the input occurred after the output spike within the same time window, then the connection was weakened. In other words, the wiring of the brain was plastic.

Markram's papers on synaptic plasticity and the microcircuitry of the neural cortex were enough to earn him a full professorship at the age of 40, but his discoveries left him restless and dissatisfied. "The first time I reconstructed an axon and saw that it was touching [dendrites belonging to other neurons] all over, I thought God, how did it decide to put a contact here and there? That's when I realised the scale and complexity of the challenge. I thought this is impossible, how am I ever going to work out how a neuron made a decision to put all those synapses there?"

This professional epiphany was mirrored by a challenge to his family life when his son Kai (Markram has five children from two marriages) was diagnosed with Asperger's, an autism spectrum disorder. The discovery came while Markram was doing his postdoctorate in Germany and led him to consult specialists all over the world and, eventually, to embark on his own study of autism. "At the time, psychiatrists were saying that autistic children had no empathy, that they were unable to form a theory of mind. But actually I found that Kai could be intensely emotional about certain things and that he seemed to know things about what you are thinking that most people don't know."

Instead, Markram and his wife Kamila, a neuroscientist, performed a series of experiments on rats to test a new theory. Using an anti-epilepsy medicine, valproic acid, that causes birth defects that mimic autism, the Markrams found that certain networks of brain cells in the acid-treated rats were much more sensitive than normal. They also found that the brain cells in autistic rats had notably more connections: their brains were "hyperconnected", enhancing the flow of information. Finally, they discovered that the amygdala, a part of the brain responsible for fear processing, had a tendency to form new connections, possibly explaining the intense fear the autistic rats developed.

The Markrams' hypothesis is that the brains of autistic individuals are similarly hyperconnected and "hyperexcitable". Rather than suffering from a deficit in perceptual abilities, autists experience the world too intensely and so take refuge by turning inward, hence his phrase "intense world theory" to explain the disorder. '"People like Kai have jacked-up brains," he says. "They have to withdraw to protect themselves."

For Markram, using predictive neuroinformatics to reverse-engineer the parts of the brain for which we have little experimental information is only the first stage of his grand scheme. The second is to marry his brain simulation with a medical informatics platform and suck up all the available data on mental diseases from public hospitals and the proprietary databases of pharmaceutical companies. This clinical data, which would include both healthy subjects and patients with widely varying conditions, could then be systematically "mined" to identify clusters of patients with similar changes in the brain.

Once these changes have been identified, Markram argues neuroscientists will be in a better position to draw up hypotheses about the underlying biological causes and test the hypotheses in his brain simulation. For instance, one patient diagnosed with schizophrenia and another diagnosed with depression might share the same mutated gene. Conversely, two patients diagnosed with schizophrenia might be found to have different gene mutations. In either case, the goal is to do away with the current classification system based on the subjective ordering of symptoms and syndromes and replace it with one that adheres more closely to biological signatures.

"We're not interested in getting more data on mental diseases," says Markram. "We are going to put all the diseases on the table and start working out mathematically how they are related to each other. There are going to be no names, just clusters. "The final stage would be to use this new biologically grounded classification system to develop new diagnostic tools and suggest strategies for drug development and treatment.

For Markram's critics, such statements strain credulity. There is no such thing as a "normal" brain, they say. Every brain is different – the serendipitous product of evolution and personal experience. Moreover, neural circuitry not only differs from one individual to another, it changes from hour to hour and day to day. Even if you could simulate one brain in all its complexity – and at Bern many of Markram's critics doubted that he would ever have sufficient computing power to run his simulation – there is no reason to believe that this model would be equivalent to a real, biological brain. But perhaps the biggest criticism of the Human Brain Project is the idea that a computer model can ever be a substitute for hard empirical research.

What neuroscience needs is diversity and a multiplicity of approaches by creative young researchers, not integrated top-down science, argues Markram's former mentor, Rodney Douglas. By betting the bank on the Human Brain Project, there is a danger that other, innovative approaches will be ignored for lack of funding. As Kevan Martin, the co-director of the Institute of Neuroinformatics in Zurich, who works closely with Douglas, pointed out in an email: "If you dump $1bn on any research some good will come of it, but its a strange way of spending taxpayers' money … so whatever your take is on Big Neuro, do not expect them to make good on all their promises to find causes, let alone cures, for any of the big neuro diseases they list in 10 years and as for new computing technologies? They are pulling your leg."

Markram's response to this is that he has never said his brain simulation is a replacement for animal experimentation; the point is to use the simulation to suggest lines of research that are likely to be most profitable. For proof his reverse-engineering methods work he points me to his Blue Brain wet lab, a short walk from his office on the other side of the Lausanne campus. The lab, which consists of a series of partitioned workstations manned by enthusiastic young researchers, is a far cry from his inner sanctum. Patch clamps and pipettes line the desks, while the walls are decorated with vivid images of rat synapses and posters demonstrating the firing patterns of different neuron types. It was here that in 2002 Markram began accumulating data on a section of the rat neocortex no larger than a pinhead. The column consists of about 30,000 neurons and thousands of synaptic pathways between neurons, only a small percentage of which have been measured experimentally.

However, by accumulating data on different cell types and the genes that encode for the expression of particular proteins and ion channels, Markram was able to model the electrical prosperities of the synapses and a form a picture of how they communicated and formed links with synapses in other parts of the column. Using his Blue Gene computer, he then ran statistical simulations to predict structures in parts of the column for which there was no experimental data. The final stage was to compare his model to the brains of real rats in his wet lab.

To date, Markram's team have simulated 100 interconnected columns and every week, Blue Gene "runs" a new model of the cortical column, incorporating the latest data from experiments, as well as the data on ion channels. This ion channel data is automatically uploaded to – a user-editable website that collates similar data from papers published across PubMed. The site is a model for the open-source team approach to sharing data that Markram would like to see become standard across neuroscience.

Markram is now busily recruiting PhDs with the aim of having his neuroinformatics and brain simulation platforms up and running within 30 months. In addition, he says he will work closely with researchers in the US, including those at the Allen Institute for Brain Science in Seattle, who are mapping the correlation between specific genes and structures in human and mouse brains and the spinal cord, and the Human Connectome Project, which is using non-invasive imaging to map the neural circuitry of human brains, integrating their data into his platform as it becomes available.

"I'm very happy Obama launched the human brain initiative," he says. "We will use any data coming out of the US and run it on our model." Indeed, Markram believes data from other brain projects will go a long way to validating his own approach. The Human Brain Project is primarily "an integration project" not a data generation project, he says. "We will do a little bit of data generation but our choices will be very strategic, studies that other people are not going to do and which are key to discovering underlying principles."

Whatever you think about Markram's vision, one thing is certain: simulating a human brain will require epic computing power, which is why one of the key sub-goals of the project is to stimulate research into an alternative approach known as neuromorphic computing.

Conventional computers are very good at complex calculations involving the parsing of large amounts of data but are very poor at performing several small tasks at once. This is because conventional computers rely on a central clock to manage data. By contrast, asynchronous signalling, the principle behind neuromorphic computing, does not use a central clock. Instead, the computer's processors are designed to spit out lots of tiny bundles of information as and when it suits them, much as a neuron spikes in response to a certain threshold of electrical activity.

One such project is SpiNNaker, short for spiking neural architecture. The idea of a team at the University of Manchester, it combines the best of analogue and digital computing and replaces neurons with custom-designed microchips powered by low wattage ARM processors designed to mimic the spiking patterns seen in the human brain. One of the limitations of using microchips is that you cannot possibly wire them with the same complexity as neurons and dendrons. But what you can do is exploit the fact that an electronic wire is much faster than a biological wire and can carry several signals sequentially, one after another. As Professor Steve Furber, who heads SpiNNaker, explains: "The solution we came up with was basically when a neuron goes 'ping' this is represented as a very small packet in an electronic communication network. We then move that packet very quickly around the system to many different places and keep the size of the packets very small compared with, say, a supercomputer."

Just as in a real brain, communications are initiated whenever a sender wants to send, and signals arrive at the receiver unheralded and must be handled, ready or not. Using this method, Furber, a division co-leader on the Human Brain Project, eventually hopes to be able to move 10 billion packets a second around his network. The problem is this will still only be equivalent to modelling about one billion neurons, or 1% of a human brain, in real time.

Furber is not the only neuromorphic engineer whose approach Markram hopes will one day pay dividends. Another is Karlheinz Meier, a physicist at the University of Heidelberg and a member of Markram's executive committee, who is building a comparable machine called Spikey. If these approaches are successful the result could be a computer capable of running a thousand times faster than a biological brain but with energy costs many times lower than a conventional supercomputer. The possibilities do not end there, however. Markram also envisages connecting his brain simulation to a robot equipped with sensors and simultaneously recording what the robot is seeing and hearing as it explores a physical environment.

By correlating the auditory and visual signals with the simulated brain activity as the robot learns about the world, researchers could then introduce distortions to the simulation – so as, say, to mimic the wiring of an autistic brain – and then hit the replay button. In theory, it should then be possible step inside a 3D hologram of the simulation and experience the world as an autist experiences it and, at the same, watch what is going on in the autist's mind.

Indeed, speaking to Markram, you get the impression that this is his real ambition and the source of his drive. "To be able to dial up everything, the colours, the sounds – that's what motivates me," he says. "To be able to step inside a simulation of my son's brain and see the world as he sees it. At the moment, I can use fMRI and ECG [electrocardiogram] to see how the brain processes information and which regions are activated during different tasks but I can't see what it is perceiving, I can't see what it sees."

To his critics, this is simply wishful thinking, the sort of statement one would expect from a sci-fi enthusiast rather than a serious neuroscientist. At TED in 2009 Markram even seemed to hold out the prospect that such a simulation might be capable of consciousness, ending his talk with the promise that if he succeeded in building his brain, then in 10 years he would "send a hologram to talk to you".

This is not the first time a scientist has made such a promise – in 1946 Alan Turing predicted that "in 30 years it would be as easy to ask a computer a question as a human being" – and since his TED talk Markram has learned to be more circumspect. When I ask him whether his simulation would be conscious for once he hesitates. "That we are not sure about," he replies eventually. "We will see a lot of neural correlates to complex cognitive behaviour but whether this is going to lead to consciousness I don't know and I don't think anybody knows."

Scientists will get a chance to grill Markram about this and other questions as his project takes shape in the months ahead. At the inaugural meeting of the Human Brain Project earlier this month, researchers from more than 80 European institutions converged on the Lausanne campus to thrash out who would contribute to what platform. Presumably, €1bn buys you more friends than enemies and Markram is likely to be inundated with suggestions for how to divvy up the research pie.

Markram is the first to admit that his 10-year time frame may be optimistic but he insists that unless we set a goal we are never going to get there. "It's like going to the moon, you need to have a mission statement." Following several years of exploratory discussions, it took a similar formal declaration of intent in 1991 to launch the Human Genome Project. At the time, it was thought the genome might be comprised as many as 140,000 genes and it could take 15 years to sequence them all. In the event, there were only 23,000 and the sequencing took only 10 years. However, to date the effort has yet to deliver the revolution in personalised medicine that was promised at the project's inception and geneticists have begun to realise that the relationship between disease and heredity is far more complex, and environment-dependent, than they first thought.

So far, there is little sign of similar hubris at the Human Brain Project, a far more complex undertaking, but perhaps for the moment Markram's ambition is precisely what is needed. "The thing that I admire about Henry is that he is a great believer that when problem reaches a certain point the only way to make progress is to industrialise the research process," says Furber, before adding that: "Whether the project will yield the big goal of giving us a clear conception of how the brain works is bound to be speculative because we have no way of knowing if we're capturing enough data or not, but it will almost certainly tell us a lot about the biology of the brain and advance computational neuroscience."

In other words, Furber concludes, it's "very unlikely not to yield a lot of useful results".

Share 0
Like 0

You are in