Donald Doherty's blog

Tag: Brain Science

  • Coding in the Brain, Paper Bloat, and the Need to Change the Way Papers are Published

    We’ve looked at some recent papers that take rate coding to task and argue that individual action potentials and their precise timing are important for signal processing in the brain. The recent paper “Sensitivity to perturbations in vivo implies high noise and suggests rate coding in cortex” (published July 1, 2010 in Nature) provides support for rate coding in the cerebral cortex. In fact, it goes further and states that the evidence rules out the importance of individual spikes and their precise timing for signal processing in the cerebral cortex.

    But what does that have to do with paper bloat? It just so happens that this six page paper has 42 pages of supplementary material associated with it. This is not unusual. At least not with the types of papers discussed in this blog. A six page paper discussed two days ago had 88 pages of supplementary material and a paper discussed last November had 177 pages of supplementary material. Each paper is becoming a book. How may we, with finite life spans, keep up?

    The nature of the journal article must change. Supplementary material is a necessity but instead of each article sitting on top of a submerged mammoth sized text it will be linked (using Semantic Web technologies) with visualization, simulations, and other high-level methods for efficiently conveying large masses of information. Naturally, interested parties will be able to drill down into text and equations but they’ll be able to absorb a lot of the information by working with the products of the study rather than reading about them.

    There is another reason to bring up the urgent need for a change in publishing practices in connection with this article. The authors base some of their work on an existing model that runs in NEURON and is available from the SenseLab ModelDB repository (Spike Initiation in Neocortical Pyramidal Neurons (Mainen et al 1995)). However, they did not publish their own models to an open repository. That’s the least they should have done to help readers evaluate their conclusions. They also should have published their physiological data to an open data repository.

    Fundamentally, the team asked if small perturbations to spiking activity in cortical networks are amplified. Here’s what they found:

    • A perturbation consisting of a single extra spike in one neuron produced approximately 28 additional spikes in its postsynaptic targets.
    • A single spike in a neuron produced a detectable increase in firing rate in the local network.
    • The observed amplification led to intrinsic, stimulus independent variations in membrane potential of the order of 2.2 to 4.5 millivolts.

    Their conclusions hinge on the idea that a well defined perturbation resulted in stimulus independent variations in membrane potential. How can the one lead to the other but, on the other hand, be independent? The authors state that the variations in membrane potential “are pure noise, and so carry no information at all.” They go on to conclude “for the brain to perform reliable computations, it must either use a rate code, or generate very large, fast depolarizing events, such as those proposed by the theory of synfire chains.” They follow this up by recording activity from layer 5 pyramidal cells in somatosensory cortex. They state that their “findings are consistent with the idea that cortex is likely to use primarily a rate code.” They came to this conclusion because they found that large, fast depolarizing events were very rare.

    The question of signal processing in the brain is fundamental so these conclusions warrant a careful look and deep consideration. Over the next couple of days I’ll post closer looks at the work reported in 48 pages of journal article and supplementary material.

    Other related blog posts:

    Neuronal Group Selection and Spike Timing Dependent Plasticity

    A Taste of Neuroscience Papers in the Future

    Spontaneously Formed Neuronal Groups Far Exceed the Number of Neurons

  • Extending Our Capabilities Through Automated Knowledge Acquisition

    Omics have been popping up everywhere. While reading the paper “Quantitative Analysis of Culture Using Millions of Digitized Books” (published January 14, 2011 in Science) I gasped at the neologism culturomics. Surely the omics trend is out of control! But wait. It occurred to me that just a couple days ago I used an omics word in a post to this blog. The word was holonomic, which was used within the context of Dr. Karl Pribram’s holonomic brain theory. Didn’t those of us working with Karl at the time he was writing his book “Brain and Perception” talk about the holonomic brain theory? And wasn’t that during the pre-omic world? Time to go to Google’s Ngram Viewer!

    Figure 1. The frequency of use of the words genomic and holonomic in English language books published from 1880 through 2008. Smoothing is set to zero to show raw data results

    A quick search for the words genomic and holonomic in books from 1880 through 2008 shows a typical lesson from history (see Figure 1 above). The words have been in use for a lot longer than I expected. Notice the little bumps as far back as the late 1890s. But how were the words used?

    First let’s consider the accuracy of the data. When I look at books cited as containing the word genomic before the 1930’s most if not all of the citations are errors. The great majority of errors are due to mistakes in optical character recognition. Many of the late 19th century mistakes are due to the French word generale but also from aenemic, economics, and Cenozoic. Some of the errors are due to wrong dates. For example, a book from 1982 may be listed as from 1882. This changes around the 1930s. For example a 1939 lecture given by by Richard Goldschmidt stated “The facts reported indicate differences between species which are on a chromosomal level and, maybe, frequently even on a genomic level.” This was published in a 1940 book titled “The Material Basis of Evolution.”

    Note: A 1-gram is a string of characters uninterrupted by a space. An n-gram is a sequence of 1-grams, such as the phrases “holonomic brain” (a 2-gram) and “the neuron doctrine” (a 3-gram). Usage frequency (y-axis in graphs by Google’s Ngram Viewer) is computed by dividing the number of instances of an n-gram in a given year by the total number of n-grams in the corpus in that year.

    The y-axis in Figure 1 shows the search word’s percentage of all the words (1-grams; see note above) in all the books published in English that are currently part of the database. Even when genomic becomes relatively common the word peaks at showing up about 0.000300% of the time. That means genomic occurs 3 ten thousandths of one percent of the time in English language books. The little bumps between 1930 and 1960 are far smaller; less than 0.00000050% or 5 ten millionths of one percent during most of the 1930s and less than 0.00000250% or 25 ten millionths of one percent (or 5 times more) during 1960. These little bumps due to genomic aren’t even detectable in the graph shown in Figure 1 but they include a high percentage of real usage (rather than errors) in ways similar to the way the word is used today. Around 1970 the use of genomic becomes detectable in Figure 1 at 0.00000700% or 7 millionths of one percent.

    Figure 2. The 2-gram holonomic brain appears in English language books beginning around the publication of “Brain and Perception” on June 1, 1991. Smoothing is set to zero to show raw data results.

    Interestingly, even though the word holonomic has remained rare, early references actually pan out as genuine rather than errors. For example, a set of 16 books containing the word holonomic was returned for between 1902 and 1905. All 16 citations were correct. On the other hand the use of the word is so rare in the corpus as to barely be detectable at less than 0.00000040% or 4 ten millionths of one percent. Holonomic was defined as “a dynamical system for which a displacement represented by arbitrary infinitesimal changes in the coordinates is in general a possible displacement” in the 1904 book “A Treatise on the Analytical Dynamics of Particles and Rigid Bodies” by Edmund Taylor Whittaker. The use was not in relation to the brain but was used in mathematical definitions of specific types of dynamical systems. It wasn’t until around the publication of “Brain and Perception” on June 1, 1991 that the 2-gram holonomic brain appears in the literature (see Figure 2 above).

    All of this points to how fun the tools and data set presented in the paper “Quantitative Analysis of Culture Using Millions of Digitized Books” can be. The paper states that over 15 million books (about 12% of all books ever published) have been digitized by Google so far. The authors carried out some cultural investigations on a subset of those data containing 5,195,769 books (about 4% of all books ever published).

    Note: Those interested in research methods and other details should download the supporting online material for this article, an 88 page pdf file, available here.

    Mass access to our published heritage is a positive development. However, even the most voracious reader may only read an extremely small percentage of published books and literature. As the authors said in the paper “If you tried to read only English-language entries from the year 2000 alone, at the reasonable pace of 200 words/min, without interruptions for food or sleep, it would take 80 years.”

    How will we, as finite beings, be able to keep up? Even within our areas of special interest? Clearly twenty-first century breakthroughs will be about extending our capabilities through automated knowledge acquisition. That’s where the Semantic Web comes in.

    Note: The full data set described in the paper is available for exploration or download at www.culturomics.org and ngrams.googlelabs.com.

    Other related blog posts:

    Sex Matters But the Brain is Like Nothing Else

  • Sex Matters But the Brain is Like Nothing Else

    My longtime friend and mentor Dr. Karl Pribram has often said to me that anything may be found in the brain. It seems that whatever the current trend is – hydraulics or computers, chaos theory or graph theory – some structure or function in the brain supports the idea in some way. It’s only a matter of time when quantum and/or optical hybrid computing machines are common and a larger audience will understand and perhaps embrace Karl’s holonomic hypothesis.

    Figure 1. The construction of a cortical anatomical network by diffusion tensor imaging. See text below for more information. From “Sex- and Brain Size–Related Small-World Structural Cortical Networks in Young Adults: A DTI Tractography Study“. By Chaogan Yan, Gaolang Gong, Jinhui Wang, Deyi Wang, Dongqiang Liu, Chaozhe Zhu, Zhang J. Chen, Alan Evans, Yufeng Zang and Yong He. Cerebral Cortex Volume 21, Number 2, February 2011.

    A currently popular topic is graph theory, which is certainly dear to my heart as core to Semantic Web technologies and forms the mathematical foundation of many other networking technologies and analyses. Naturally, graph theory is currently being used while investigating many different things including the central nervous system.

    The authors of the new paper “Sex- and Brain Size–Related Small-World Structural Cortical Networks in Young Adults: A DTI Tractography Study” (published February 2011 in Cerebral Cortex) hypothesized that there are sex and brain size related differences in the patterns of anatomical connectivity in the human brain. To test their hypothesis, the research team used diffusion tensor imaging (DTI) techniques on 72 healthy young human adults to construct interregional connectivity for each participant and calculate topological parameters using graph-theoretical approaches. They then investigated the association of interregional connectivity and topological parameters with sex and brain size.

    The research team derived the cerebral cortical network connectivity of each individual using a multi-stepped process, which is outlined visually in Figure 1 above. The structural image of a brain was first transformed into what is called diffusion tensor imaging native space (a). Next the image was segmented into grey matter, white matter, and cerebrospinal fluid (b; left image). The automated anatomical labeling template (b; right image) was applied to an individual’s segmented diffusion tensor imaging native space brain image (c). In parallel, each individual’s white matter fibers were reconstructed in the whole brain by using diffusion tensor imaging deterministic tractography (d).

    The result is a brain mapped into 39 cerebral cortical regions within each hemisphere and associated with a connection matrix weighted according to the number of fibers connecting the regions. At bottom (e) left is a connection matrix color coded for the fiber density between each pair of regions. At bottom middle and right are views of the brain showing connections as graphs with nodes placed at the center of mass for each of the 39 brain regions per hemisphere. Edge thicknesses are coded according to the number of fibers connecting the regions.

    Consistent with previous studies, the research team found that women’s brains were significantly smaller than those of men. The team found that the difference in brain size between males and females remained after correcting for height. Unfortunately, they didn’t report if the difference remains significant after correcting for overall body mass. Also consistent with previous research results is their finding that the anatomical networks of the human brain have “small-world network” characteristics with relatively greater local interconnectivity and an emphasis on the shorter connections between regions.

    This study goes beyond earlier studies by showing that females have greater local clustering in cortical anatomical networks as compared with males, suggesting higher local network efficiency in the female brain. They also found that brain size is significantly and negatively correlated with local clustering, suggesting that smaller human brains are more efficient in local information transfer. Interestingly, they found that the brain size effect on local efficiency was not significant in males.

    The brain is like many things. Probably because the brain is really completely unlike anything humans have ever built or understood. Put another way, Newtonian physics is an approximation (it’s “like”) relativity physics within a particular scale of time and space. The brain encompasses so much structural and functional complexity that there are a lot of things “like” in the brain. It is fun and sometimes even useful to say that a certain system in the brain is like a telephone switchboard, a computer, or a social network. However, we must keep in mind that the brain is really not the same as anything else we know.