Song L.,Georgia Institute of Technology |
Smola A.,Yahoo! |
Gretton A.,Gatsby Computational Neuroscience Unit |
Gretton A.,Intelligent Group |
And 3 more authors.
Journal of Machine Learning Research | Year: 2012
We introduce a framework for feature selection based on dependence maximization between the selected features and the labels of an estimation problem, using the Hilbert-Schmidt Independence Criterion. The key idea is that good features should be highly dependent on the labels. Our approach leads to a greedy procedure for feature selection. We show that a number of existing feature selectors are special cases of this framework. Experiments on both artificial and real-world data show that our feature selector works well in practice. © 2012 Le Song, Alex Smola, Arthur Gretton, Justin Bedo and Karsten Borgwardt.
Ordinarily, of course a fly does not shimmer green. Here, researchers - with the help of genetic tricks - succeeded in making a muscle protein glow, which they can then locate under a fluorescence microscope. Credit: MPI for Biochemistry The human genome codes for more than 20,000 different proteins, however the molecular role for many of these proteins is not known. As most proteins are conserved from fly to humans, understanding the molecular role of a protein in flies can be the first step towards a therapy against a variety of human diseases that are often caused by aberrantly behaving proteins. A consortium of scientists from the Max Planck Institutes of Biochemistry in Martinsried and Molecular Cell Biology and Genetics in Dresden, and the National Centre for Biological Sciences (NCBS) in Bangalore have now reached a milestone towards understanding the function of these proteins by using the fruit fly. The human body is built by many hundreds of different cell types; each one has a very particular function in the body. Red blood cells transport oxygen, nerve cells exchange signals and muscle cells generate mechanical forces. The majority of cellular functions is produced by the action of 20,000-25,000 proteins coded in the human genome. Although sequencing and annotation of human genome were completed in 2004, to date the function of many thousands of these proteins is still mysterious. It is often unknown, which cell types produce which proteins, and particularly where these proteins are located within the cells. Are they in the nucleus or within membranous vesicles, are they within neuronal dendrites or synapses, or are they within the contractile machinery of muscles? Protein localisation is an important piece of information, as it is the first step towards identifying a molecular function for a protein. Unravelling the function of a protein is often started in simpler model organisms such as worms or flies. Like humans, fruit flies have muscles, neurons, oocytes, sperm and many other essential cells types. The fly genome contains about 13,000 protein coding genes, which are responsible for building and maintaining all fly organs. Importantly, many of these proteins are very similar to the human proteins, thus studying a protein in flies will teach us about its role in the human body. To boost these protein studies onto a systematic level, groups headed by Frank Schnorrer at the Max-Planck Institute in Martinsried, Pavel Tomancak and Mihail Sarov at the Max-Planck Institute in Dresden and K VijayRaghavan at the NCBS in Bangalore have generated a large resource for visualizing proteins in Drosophila melanogaster. By using modern molecular biology tricks, the scientists have attached a green fluorescent protein (GFP) tag to 10,000 of these protein coding genes in the test tube. Each tagged gene can then be re-introduced into the fly genome as a 'transgene', creating the fly 'TransgeneOme'. "Together, we thus far generated 880 different fly strains, each of which expresses a different fluorescently tagged protein", explains Frank Schnorrer, "these proteins can then be observed by fluorescent video microscopy in various cell types of the developing fruit fly". For more than 200 proteins, the scientists documented where they are located during fly development, starting with an oocyte that develops into an embryo and finally into the mature fly. The Tomancak group used the so-called light sheet microscopy to film how proteins emerge in cells of the embryo during the first day of its development. The Schnorrer group used this resource to study the localisation of proteins in muscles. As in human skeletal muscles, fly muscles contain complex mini-machines called sarcomeres that produce the mechanical forces enabling animal movements. "We have looked so far at only 200 of these transgenic lines. The future challenge lies in systematically imaging the localization of these proteins in many fly tissues and this is best achieved by involving the powerful Drosophila research community" predicts Pavel Tomancak. The resource will have enormous impact on the understanding of not only fly biology but also on the understanding of protein function in the different human cell types." More information: Mihail Sarov et al. A genome-wide resource for the analysis of protein localisation in , eLife (2016). DOI: 10.7554/eLife.12068
The method of taking these pictures is a collaborative creation that involved Kansas State University researchers Artem Rudenko and Daniel Rolles, both assistant professors of physics. The movies help scientists understand interactions of intense laser light with matter. But even more importantly, these experiments lead the way to filming various processes that involve ultrafast dynamics of microscopic samples, such as the formation of aerosols—which play a major role in climate models—or laser-driven fusion. "We can create a real movie of the microworld," Rudenko said. "The key development is that now we can take sequences of pictures on the nanoscale." Rudenko and Rolles—both affiliated with the university's James R. Macdonald Laboratory—collaborated with researchers at SLAC National Accelerator Laboratory at Stanford University, Argonne National Laboratory and the Max Planck Institutes in Germany. Their publication, "Femtosecond and nanometre visualization of structural dynamics in superheated nanoparticles," appears in Nature Photonics. In this work, the collaboration used intense lasers to heat xenon nanoscale clusters and then took a series of X-ray pictures to show what happened to the particles. The picture series became a movie of how these objects move at the level of femtoseconds, which are one-millionth of a billionth of a second. "What makes nano so interesting is that the behavior for many things changes when you get to the nanoscale," Rolles said. "Nano-objects bridge the gap between bulk matter and individual atoms or molecules. This research helps us as we try to understand the behavior of nano-objects and how they change shape and properties within extremely short times." The pictures of the nanoparticles cannot be taken with normal optical light, but must be taken with X-rays because X-ray light has nanometer wavelengths that enable researchers to view nanoscale objects, Rolles said. The light wavelength must match the size of the object. To take the pictures, the researchers needed two ingredients: very short X-ray pulses and very powerful X-ray pulses. The Linac Coherent Light Source at SLAC provided those two ingredients, and Rudenko and Rolles traveled to California to use this machine to take the perfect pictures. The photo-taking method and the pictures it produces have numerous applications in physics and chemistry, Rolles said. The method is also valuable for visualizing laser interactions with nanoparticles and for the rapidly developing field of nanoplasmonics, in which the properties of nanoparticles are manipulated with intense light fields. This may help to build next-generation electronics. "Light-driven electronics can be much faster than conventional electronics because the key processes will be driven by light, which can be extremely fast," Rudenko said. "This research has big potential for optoelectronics, but in order to improve technology, we need to know how a laser drives those nanoparticles. The movie-making technology is an important step in this direction." Rudenko and Rolles are continuing to improve the moviemaking process. In collaboration with the university's soft matter physics group, they have extended the range of samples, which can be put into the X-ray machine and now can produce movies of gold and silica nanoparticles. Explore further: Dual camera smartphones – the missing link that will bring augmented reality into the mainstream More information: Tais Gorkhover et al. Femtosecond and nanometre visualization of structural dynamics in superheated nanoparticles, Nature Photonics (2016). DOI: 10.1038/nphoton.2015.264
Camps-Valls G.,University of Valencia |
Shervashidze N.,Max Planck Institutes |
Borgwardt K.M.,Max Planck Institutes
IEEE Geoscience and Remote Sensing Letters | Year: 2010
This letter presents a graph kernel for spatio-spectral remote sensing image classification with support vector machines (SVMs). The method considers higher order relations in the neighborhood (beyond pairwise spatial relations) to iteratively compute a kernel matrix for SVM learning. The proposed kernel is easy to compute and constitutes a powerful alternative to existing approaches. The capabilities of the method are illustrated in several multi- and hyperspectral remote sensing images acquired over both urban and agricultural areas. © 2006 IEEE.
Kam-Thong T.,Max Planck Institute of Psychiatry |
Czamara D.,Max Planck Institute of Psychiatry |
Tsuda K.,Max Planck Institute for Biological Cybernetics |
Tsuda K.,Japan National Institute of Advanced Industrial Science and Technology |
And 14 more authors.
European Journal of Human Genetics | Year: 2011
Detection of epistatic interaction between loci has been postulated to provide a more in-depth understanding of the complex biological and biochemical pathways underlying human diseases. Studying the interaction between two loci is the natural progression following traditional and well-established single locus analysis. However, the added costs and time duration required for the computation involved have thus far deterred researchers from pursuing a genome-wide analysis of epistasis. In this paper, we propose a method allowing such analysis to be conducted very rapidly. The method, dubbed EPIBLASTER, is applicable to case-control studies and consists of a two-step process in which the difference in Pearson's correlation coefficients is computed between controls and cases across all possible SNP pairs as an indication of significant interaction warranting further analysis. For the subset of interactions deemed potentially significant, a second-stage analysis is performed using the likelihood ratio test from the logistic regression to obtain the P-value for the estimated coefficients of the individual effects and the interaction term. The algorithm is implemented using the parallel computational capability of commercially available graphical processing units to greatly reduce the computation time involved. In the current setup and example data sets (211 cases, 222 controls, 299468 SNPs; and 601 cases, 825 controls, 291095 SNPs), this coefficient evaluation stage can be completed in roughly 1 day. Our method allows for exhaustive and rapid detection of significant SNP pair interactions without imposing significant marginal effects of the single loci involved in the pair. © 2011 Macmillan Publishers Limited All rights reserved.