The Price of Knowledge

It’s not exactly neuroscience, but I think that open access publishing is of general enough interest for readers of this blog. This article appeared in Issue 2 of EuSci, the University of Edinburgh’s science magazine. Available here.

---

I’ve set up a computer software company with a twist. Instead of going the usual route and hiring a team of programmers to develop my new applications, I solicit the public for them. On top of that, I don’t offer a penny. Regardless, people race to give me their brightest ideas before their friends can beat them to it. This allows me to cherry-pick the products I think will make the biggest impact on the marketplace. Of course, even the best submissions need a bit of polishing before they’re fit for general distribution. No problem. I just get a few of their amateur programming buddies to do the debugging for me – free of charge, obviously. All that’s left to do is package the software up and sell it right back to the masses. Easy money.

Of course the company I’ve just described is entirely fictional. Its business model, however, is not. It is exactly the strategy employed by many publishers of academic journals. The millions of global university academics, industrial researchers and independent scholars generate almost all of the content in today’s academic journals. Candidate papers are usually submitted to a journal editor who filters out the junk. Typically, the work is then outsourced back to academia for peer-reviewing. If a paper is still deemed up to scratch, it can be formatted and issued as part of a periodical publication. This is a highly profitable business. Elsevier, the science and medical wing of publishing giant Reed Elsevier, currently prints over 2000 scholarly journals and posted an adjusted operating profit of £477m in 2007 alone.

So what is the problem? Journal publishers provide a necessary service, work hard to maintain high quality publications, and so perhaps should be adequately rewarded for it. Besides, no-one loses out in this arrangement. Subscription fees are usually paid by university libraries, not the scholars themselves. Accessing and indexing journal papers has never been easier, thanks to the print-to-digital switch. And nobody else cares about the esoteric topics covered in these publications anyway. Do they?

The open access movement would argue differently. To start with, the majority of people on Earth live in developing countries where funding for academic study may be minimal. Many institutes in these countries cannot afford the ever-increasing subscription fees demanded by commercial publishers. Consequently, they can hold only small numbers of current journals. No matter how talented a scholar is, producing world class research is an impossibility if you cannot keep up to date with work in your field.

Perhaps more surprisingly, the same issue is also becoming a problem in westernised countries. Science journal prices rose 30-40% on average from 2004 to 2008, while inflation over the same period was less than 15%. As libraries’ funding rarely increases faster than inflation, many institutions are being forced to reduce the size of their catalogue. The ‘serials crisis’ has become so bad that several influential institutions have staged protests. In 2004, US universities Harvard and Cornell both cancelled their subscriptions to one of Elsevier’s bundle packages containing more than 900 journals, instead choosing to hand-pick individual titles. Bundling is a tactic common to many publishers where many journals are grouped together as an all-or-nothing package. Although the packages are much cheaper than the sum of their parts, they also often contain many irrelevant journals which some institutions might not need.

Open access publishing is an alternative model to the dominant commercial approach. Its primary philosophy is that access to knowledge should be a right to all, rather than a commodity to be bought and sold. This should be doubly true given that the vast majority of scholarly research is funded by public money in the first instance. The ideal was spelled out in John Willinsky’s 2005 book on the subject,The Access Principle: “a commitment to the value and quality of research carries with it a responsibility to extend the circulation of such work as far as possible and ideally to all who are interested in it and all who might profit by it”. Naturally, Willinsky’s book is freely available online.

No matter how noble a cause freeing up journal access is, someone still has to pick up the bill. Editors, printers and website staff all have mortgages to pay. The most straightforward solution proposed so far has been to charge the authors a publishing fee. For example, the Public Library of Science’s flagship journal PLoS Biology currently charges $2850 for an accepted publication. For this one-off fee, the journal covers peer-review, journal production, online hosting and archiving. The articles are freely available to everyone with internet access. Through this system (and private donations) PLoS successfully funds their entire not-for-profit organisation.

Although a $1000+ charge might seem exorbitant, it is important to remember that this is often insignificant when compared to the salaries, equipment and housing costs that went into producing the research. Even so, PLoS have pledged to waive or discount the fee for those who cannot afford it. Funding bodies like the Wellcome Trust, the EC and NIH in the UK, Europe and USA respectively have also pitched in by pledging to provide additional funds if necessary to allow research to cover open-access fees. In May 2008, NIH even went a step further by making open access a mandatory requirement for any publication at least partially funded by their money. Articles must be submitted to their online repository PubMed Central within 12 months of the official date of publication.

Commercial publishers are also beginning to realise that open access is here to stay. Many are toying with their business models in the search for solutions that improve access but maintain financial viability. One such hybrid model is the opt-in policy adopted by the US journal PNAS, commercial publishers Springer and others. Here, authors can choose to pay a fee to have their articles openly accessible, even when the other articles in the same journal remain behind subscription. Perhaps an even more convincing hint that commercial publishers are taking the open access business model seriously is the recent purchase of open-access publishers BioMed Central by Springer.

Another increasingly popular choice – also adopted by PNAS – is delayed open access. Articles are initially published behind subscription, but then made freely available after a 6 or 12-month wait. This allows the journal to retain institutional subscriptions because of academia’s interest in new research, while opening up older content to a wider audience.

Academics themselves can also do a lot to promote open access. The obvious first step is to simply publish new research in open access journals, or in journals that offer a paid open access choice. This can be to the author’s benefit, as studies have suggested that freely available articles may have a higher impact than closed ones.

Another straightforward option is to publicly archive all published work. Apart from a few restrictions, this is completely allowed by a surprising number of journals – including Science and Nature – and actually mandated by many funding bodies. The SHERPA organisation maintains an excellent website which details individual publisher and funding body open access policies.

Many academics simply archive their work on personal websites, but other options exist. Some disciplines already have popular public archives, such as the physics repository arXiv.org. Most papers in this field are posted on ‘the archive’ well before being accepted in a journal, with no apparent detriment to the publishers. Many academic institutions also maintain their own archiving facilities. Here in the University of Edinburgh, staff and students can self-archive their own work in the Edinburgh Research Archive. The technophobic can also get library staff to deposit work on their behalf.

Best guesses at the total number of current scholarly journals are in the tens of thousands. Of these, more than 3500 are fully open access (a list is maintained at doaj.org). This means that a small, but growing fraction of scholarly work is now freely available to anyone with a connection to the web. In the wikipedia age we have no shortage of instantly accessible information, but sadly, facts and figures are not always backed by expert opinion. The open access movement aims to remedy this by making scholarly knowledge available and accessible to all who wish to find it.

Computational Neuroscience: modeling the mind

I wrote this article for issue 1 of the University of Edinburgh science magazine (available in pdf format from here). It is a synopsis of some of the basic ideas and methods used by computational neuroscientists to study the brain. Of course it is probably a biased account from a niche viewpoint, but what isn’t in this world. My main hope is that in future I can just direct people to this article instead of fumbling out an incoherent non-explanation. Comments and criticism very welcome.

---

When people gesture while talking, it is usually for one of two reasons. If they are sure of themselves, hand movements can bring emotion and conviction to the words. If they are not so sure, they might be using the hand waving to convince you that they are right in general, and that the details are not important anyway. The latter is often the case when people talk about the brain.

It is not that what goes on beneath our scalp is a complete mystery. Modern neuroscience began over 100 years ago, when pioneering neuroanatomists began to unearth the basic architecture of the central nervous system. Since then, the field has grown significantly and today its largest conference, the Society for Neuroscience Annual Meeting, attracts over 30,000 attendees. Thanks to the popularity of the subject and the seemingly never-ending technical advances, scientists are now churning out masses of data at every level of analysis. Every day, they fill databases with gene sequences and protein interactions, map out networks of nerve cells, and even record the differing roles of each brain region in behaviour.

So why all the uncertainty? If all these experts are working so hard, why have we not found a cure for Alzheimer’s, understood how you can tell a cat from a dog in a split second, or explained why I feel like a person, and not just “a pack of neurons”, as suggested by Francis Crick? The answer lies in the sheer complexity of the brain. A human adult has about 100 billion neurons inside their head, all working away at their own little chores. Scientists have extensive knowledge about the different cell types, their make-up, how they are wired together, and ideas about what most of them are doing. But the leap from this to something that can do a crossword puzzle is a big one. It is not an impossible problem, exactly. Just a hard one.

Like other scientific disciplines before it, neuroscience is now reaching a stage where enough facts are known to start building general, and maybe even mathematical theories about how it all works. Computational neuroscience is the field that develops and tests these theories. That is not to say that experiments will ever become obsolete. Even relatively mature fields, such as physics, need the constant challenge of real life experiments to show who is right and who is wrong. The aim of computational neuroscience right now is to gather existing experimental data, try to fit it together in some coherent way, and go on to make suggestions and predictions for future experiments.

So how does someone tapping away at a computer in a dusty old office study the brain? They do it by trying to build theoretical models. A good example of this is a popular method called compartmental modelling, often used to examine the behaviour of single neurons (nerve cells). Since each neuron in our brain computes and transmits information using electrical signals, it is possible to think of each of them as a small, individual electrical circuit, made from the same basic elements – resistors, capacitors and the like – that control your mobile phone. In principle I could go, soldering iron in hand, and physically make a model neuron with these building blocks. Some people do. The downside is that it is a very time and resource-sapping process. It’s much easier to build a virtual circuit on your computer.

With enough constraining experimental data, these types of single-cell models can become quite detailed and include ion channels, complex molecular interactions and the varying shapes of real neurons. Once you have set up your model, you can test your hand-waving ideas explicitly and see if they fit together in a logical way. Another great advantage to this approach is that you can also do experiments on your virtual cell that are difficult, impossible or even immoral in real life, keeping animal rights activists happy in the process.

With enough data, these models can make specific statements about the real world. For example, an elegant study by Agmon-Snir and colleagues (Nature, 1998) looked at time-difference detecting neurons in the auditory brainstem. Imagine you are watching a tennis match from the stands. The grunting noise from the player on your left side will reach your left ear slightly earlier in time than your right ear. The further to the left the player is, the bigger the time difference. Your brain uses this information to tell which direction a sound came from. One puzzle was why, among the neurons involved, the ones that respond to higher-frequency sounds are smaller than those that deal with lower-frequency sounds. Agmon-Snir and colleagues created realistic computational models of these cells and showed that the higher frequency sound signals are optimally handled by neurons with shorter branches, because of noisy signal transmission.

Of course this single-cell example looks at just one of the many levels at which the brain could be studied. David Marr, an influential early theorist, defined three levels at which we can analyse a computational system such as the brain: the computational level, the algorithmic level and the implementation level. The first level identifies the computations that are to be performed. An example in the visual system would be motion detection. The second level determines the strategy used to perform this task. A computer programmer would call this the choice of algorithm. The third level looks at the physical implementation of this strategy, which in the case of the brain is the network of neurons. To a certain extent, experimental neuroscience has focused on this last ‘nuts-and-bolts’ level. Theoretical neuroscientists, however, have been working on all three levels; everything from the detailed biophysics of ion channels to more abstract full-brain models.

After defining a problem at a certain level, the theorist must design the model, taking into account several factors. Firstly, a model that is too detailed can be just as difficult to draw conclusions from as the real thing, which would render it mostly useless. To paraphrase Einstein, a model should include just enough detail to explore the question at hand and no more. Secondly, in many cases little experimental data is available to con- strain the model and ensure it reflects reality. The data must also be of high enough quality; substandard data will give you substandard results (a principle known among computer programmers as GIGO: Garbage In, Garbage Out). Thirdly, even with the extraordinary speed of modern computers, some simulations can take days or even weeks to run. For this reason, modellers may not want to include all the details, and may instead use approximations. Fortunately, computer processing power continues to increase every year. Many modern studies are based on ideas that others had decades ago but simply lacked the computational resources to implement at the time.

Neuroscience used to be a divided field, with the experimentalists complaining that the theorists were fiddling around with abstract ideas that would never work in a real brain, and the theorists criticising the experimentalists for filling the literature with reams of boring data. These divisions are rapidly fading. Many researchers are realising that steady progress will require a two-way flow of ideas. Even more scientists are actively blurring the lines by adapting methods from both approaches. This inter-disciplinary outlook will ensure that exciting times lie ahead for our understanding of the brain and, in many ways, of ourselves.

---

Refs:

• The role of dendrites in auditory coincidence detection.
  H Agmon-Snir, C E Carr, J Rinzel.
  Nature (1998) 393 (6682) 268-72
  PMID: 9607764
• Computational neuroscience.
  T J Sejnowski, C Koch, P S Churchland.
  Science (1988) 241 (4871), 1299-306
  PMID: 3045969

The Blue Brain shows gamma oscillations

Cortical oscillations and synchrony have long been touted as candidate mechanisms to solve the ‘binding problem’ in theoretical neuroscience: when we examine the world around us, how do our brains group multiple parts of the same object together into a coherent whole? A simple example is the cat standing behind a fence. Even though whole segments of the kitty might be blocked off from our view, we still perceive it as a single object. This happens even when the segments of the visual scene are too far apart to be seen by overlapping cells in the retina – so the information must be ‘bound’ somewhere else in the brain.

Experimental evidence implicating oscillations in this process was first found by Wolf Singer’s lab in Germany in the late 1980s (Gray et al, 1989). They reported that spatially seperated neurons in cat visual cortex mostly fired at the same time when the cat was presented with moving bars of light, as long as the neurons both preferred bars of the same orientation and were aligned in the same direction as the moving bar stimulus. This might seem like a banal result, but it hinted for the first time that neurons in the neocortex might encode information in the exact timing of their spikes (relative to some external osciallation), rather than just through their firing rate over longer time periods. In this way, spatially seperated neurons might somehow co-ordinate their firing patterns to become part of the same neuronal ensemble, and maybe represent specific features of the outside world.

Following this discovery, a sustained experimental and theoretical scientific interest has resulted in a huge library of data exploring the theory, and even implicating oscillations in attention and consciousness. Many, many debates and arguments have ensued over the origins of these oscillations and whether or not they are really used by the brain to code information. I am not going to attempt to dip my toe into this ocean here. If you’re interested, I suggest looking up the experimental work both of Wolf Singer and his former student Pascal Fries, who now runs his own lab in Nijmegen, Netherlands. BU’s Nancy Kopell has been a driving force on the more theoretical aspects of oscillations.

Despite the mountain of work on this topic, a real mechanistic description of these oscillations has yet to be demonstrated in a realistic computational model of the brain. The Blue Brain project – that other big science experiment in Switzerland – might finally make the link. Earlier this week at the inagural INCF conference on Neuroinformatics, Henry Markam reported that a recent modification to their detailed simulation of a rat cortical column produced persistent oscillatory activity in the gamma frequency band (roughly 40-80Hz).

This is significant, because the model wasn’t designed in any way to produce this behaviour. It simply emerged after setting up the cortical column of 10,000 cells with realisitic connectivity patterns and electrophysiological properties. As far as I understood, they simply stimulated layer IV and watched a wave of activity build up, propagate throughout the column via layer II/III and initiate gamma ocillatory activity in layer V. This behaviour only emerged following one of their weekly updates to the simulation. Markram wouldn’t say exactly what changes they made, unsurprisingly enough. Expect a publication forthcoming.

We could argue all day about the Blue Brain project and its significance. Many people (especially other experts) do. I had seen several presentations from the BB team before, and I suppose that I had kind of made up my mind that it was probably going to be a useful logistical excercise which would generate new tools for neural data sharing and analysis, but ultimately hopeless in helping us to understand the brain. As Markram himself admitted, the whole model is “half-baked”, and the experiment was done without a hypothesis. There are so many gaps in our knowledge (e.g. plasticity rules, dendritic ion channel distributions, neuromodulation) that the whole endeavour seemed to me to be a waste of time. But after seeing Markram’s talk last Monday, I am sold. This is mainly for two reasons:

• Of course the current model isn’t perfect. But it is a first step on the long road to a biologically realistic large-scale model of the brain. This will be a slow, iterative process.
• Despite the astronomical number of parameters, the model is actually fairly well constrained by biology. It’s in the right ball park. There are many phenomena you can reproduce in any abstract network model which wouldn’t work in the Blue Brain (or a real brain for that matter). As Markram says, why explore all of theoretical parameter space when we can focus on the biogically relevant subregion within it?

You can watch Markram’s talk here. The gamma oscillations are demonstrated 47 minutes in.

Refs:

• Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties.
  C M Gray, P König, A K Engel, W Singer.
  Nature (1989) vol. 338 (6213) pp. 334-7
  PMID: 2922061

Nothing but networks of neurons

Why are elite sport players so much better at their game than you or I? A professional basketballer would probably make eight or nine free throws from every ten, while I’d be happy with five. Why the difference?

There are a variety of reasons. The pro has probably developed (or been born with) muscle areas which I rarely use. Also, he or she has learned, through intense practice, how to control these muscles very precisely. This is how a golfer can put the ball within a couple of metres of the flag from halfway down the fairway. Despite our instinctive notions to the contrary, this fine motor control does not originate from the muscles themselves, but is orchestrated by the brain – where, of course, all learning occurs.

So, apart from being able to effectively co-ordinate their finely tuned muscles, do the top athletes have any other advantages over the rest of us? A paper published today in Nature Neuroscience suggests that they do. Aglioti et al investigated the ability of elite basketball players to anticipate whether a shot was heading for the basket or not. It turned out that these athletes can not only perform this task significantly better than expert watchers (sports journalists and coaches) and lay persons, but that their muscles activated differently for shots going in or out of the hoop, even though they were just sitting still in a chair for the duration of the experiment. And it doesn’t take an expert to see that the ability to predict the future would boost your chances of the catching that next rebound.

Intriguingly, the elite athletes’ advantage seems to stem not from being better able to visualise the path of the ball through the air, but from a superior ability to read body movements before the ball has left the player’s hands. The task was to watch a video clip of people shooting baskets. The clips were cut short at varying times before the ball reached the basket. Unsurprisingly, seeing more of the clip improved all groups’ (players, expert watchers and novices) ability to guess whether the ball finished in or out of the basket. But for the clips that were cut just at or before the player in the video released the ball (781ms in the graph), the elite basketball players were getting it right 70% of the time, compared to about 40% for the other groups. Here’s a glimpse of the data for the curious:

It would be interesting to see if the same phenomenon was found in top sports people who play games where anticipation is not a key skill, like golf or swimming. If you are interested, I recommend reading the full paper, especially for more references. I’m not an expert on the topic, but this action-perception debate and the mirror system theory is really fascinating.

Ref:
Salvatore M Aglioti, Paola Cesari, Michela Romani, Cosimo Urgesi (2008). Action anticipation and motor resonance in elite basketball players Nature Neuroscience, 11 (9), 1109-1116 DOI: 10.1038/nn.2182

The grid cell code

In 2005, Torkel Hafting and Marianne Fyhn permanently changed the neuroscience of spatial navigation by discovering the grid cell. In their seminal Nature paper, the Moser lab reported that certain cells in the rat medial entorhinal cortex (MEC) fire action potentials only when the animal occupies certain locations in space. When you map at these locations from above, they look suspiciously like a 2-D triangular grid which appears to tile any sized area. The grid persists when the rat is placed into a new environment, and even when it is kept in the dark. It is an internal, metric representation of space. To me at least, this is stunning.

The image below is a beautiful example recording. On the left, the black trace is the path of the rat over a few minutes roaming inside a 1m by 1m box, and the red dots are where the cell fired. The right image is a colour-coded map of the firing rate of the same cell. Note the triangular grid pattern. I know it’s not perfect, but in biology this is about as orderly as things get.

(I stole this image from the excellent Scholarpedia article on grid cells by Mr and Mrs Moser. I highly recommend it, along with their recent Annual Review of Neuroscience paper, for great overviews and more references.)

Along with simply reporting their discovery, the Moser group went on to describe the properties of these grids in some detail. They found that, in general, neighbouring grid cells don’t have overlapping firing fields. Although the grids of nearby neurons do usually have the same spacing and rotation, their firing patterns are typically shifted with respect to each other, in a seemingly random way. This means that if you were to collect the output from enough neighboring grid cells, then at any given moment in time and no matter where the rat is in space, some, but not all of these cells will be firing. You (or a downstream neural circuit) could potentially use this information to track the rat’s relative position within this ‘cognitive map’.

Although the repeating grid pattern might be considered aesthetically pleasing, one problem with this type of coding scheme is that it provides only ambiguous information about the rat’s position. Each neuron fires at a (presumably) infinite number of locations in space, so how could this code ever be used to represent a unique position? One clue to a possible solution lay in the topographic relationship the Mosers uncovered between the spacing of a given cell’s grid pattern and its physical location in the MEC. More dorsal cells (toward the top of the head) were found to have progressively smaller grid spacing than more ventral cells (away from the top of the head). When grid spacing is plotted against dorso-ventral location it looks roughly proportional. See the below figure for some examples (taken from McNaughton et al, 2006). The more dorsal cells (toward the upper part of the image) have smaller grid spacing.

Hafting et al suggested that a unique location could be represented by “integrating over grids with different spacing and orientation”. If you look at the above image you can see that there are very few locations in space where all three cells would be simultaneously firing. If you now imagine looking at the firing fields of 50 or 100 or 500 of these cells there might be only one or two locations in a very big area where all the cells would be firing together. In this way a population of grid cells could code for a single specific location, even though their individual activity is ambiguous. The Mosers and other groups have since elaborated on this idea to suggest that the place cells found in the hippocampus are formed by summing the output from many grid cells (see Solstad et al, 2006).

So why this unusual coding scheme? Why has evolution favoured this surprising triangular grid representation? No-one really knows for sure.

Refs:

• Microstructure of a spatial map in the entorhinal cortex.
  Torkel Hafting, Marianne Fyhn, Sturla Molden, May-Britt Moser, Edvard I Moser.
  Nature (2005) 436 (7052), 801-6
  PMID: 15965463
• Grid Cells.
  Edvard Moser, May-Britt Moser.
  Scholarpedia (2007), 2(7):3394

  http://www.scholarpedia.org/article/Grid_cells

• Place Cells, Grid Cells, and the Brain’s Spatial Representation System.
  E Moser, E Kropff, M Moser.
  Annu Rev Neurosci (2008) 31, 69-89
  PMID: 18284371
• Path integration and the neural basis of the ‘cognitive map’.
  Bruce L McNaughton, Francesco P Battaglia, Ole Jensen, Edvard I Moser, May-Britt Moser.
  Nat Rev Neurosci (2006) 7 (8), 663-78
  PMID: 16858394
• From grid cells to place cells: a mathematical model.
  Trygve Solstad, Edvard I Moser, Gaute T Einevoll.
  Hippocampus (2006) 16 (12), 1026-31
  PMID: 17094145