2007 Winter School in
Mathematical & Computational Biology
25 - 29 June 2007Queensland Bioscience Precinct
Brisbane, QLD
PROGRAM
Hosted by:ARC Centre in Bioinformatics
andInstitute for Molecular Bioscience
Sponsors:The MathWorks Australia
Queensland Cyber Infrastructre FoundationSGI
ARC Centre in Bioinformatics
CONTENTS Timetable ……………………………………………………………………………………………………… page 1 Abstracts ………………………………………………………………………………………………………. page 4
TIMETABLE
1 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Monday 25 June 2007 Bio-image analysis, quantification and
classification 08:30 a.m. – 09:30 a.m.
Registration and morning tea
09:30 a.m. – 09:40 a.m.
Opening and Welcome
Prof John Mattick
09:40 a.m. – 10:10 a.m.
An introduction to cellular imaging techniques
Dr Rohan Teasdale
10:10 a.m. – 11:40 a.m.
Automated interpretation of subcellular patterns in microscope images: bioimage informatics for systems biology
Dr Estelle Glory
11:40 a.m. – 01:15 p.m .
Lunch
01:15 p.m. – 02:15 p.m. On models and algorithms for analysis of biological
images
A/Prof Tuan Pham
02:15 p.m. – 03:15 p.m.
High content cellular screening for drug discovery
Ms Leanne Bischof
03:15 p.m. – 03:45 p.m.
Afternoon tea
03:45 p.m. – 04:45 p.m.
Reconstructing the mammalian cell
Oliver Cairncross Andrew Noske
05:30 p.m.
Welcome BBQ at rooftop, QBP (Sponsored by SGI)
Tuesday 26 June 2007
Modelling and simulation of cellular processes
09:00 a.m. – 10:00 a.m.
The role of noise in cellular processes
Prof Kevin Burrage
10:00 a.m. – 10:30 a.m.
Morning tea
10:30 a.m. – 11:30 p.m.
Virtual heart disease
Dr Edmund Crampin
12:00 p.m. – 01:30 p.m.
Lunch
01:30 p.m. – 02:30 p.m.
The numerical solution of the master equations in molecular biology – how, why and for what
Dr Markus Hegland
02:30 p.m. – 03:00 p.m.
Afternoon tea
TIMETABLE
2 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
03:00 p.m. – 03:45 p.m.
Stochastic modeling and simulation of cellular processes with delays
Dr André Leier
03:45 p.m. – 04:30 p.m.
Modelling and simulation in computational biology Mr Paul Taylor
04:30 p.m. – 05:00 p.m.
Discussion session
Wednesday 27 June 2007 Prediction and modelling of protein
structure and dynamics 09:00 a.m. – 09:45 a.m.
Stimulating biomolecular systems I
Prof Alan Mark
09:45 a.m. – 10:30 a.m.
Determination of protein complexes and multi-domain proteins using a combination of experiments and calculations
Dr Thomas Huber
10:30 a.m. – 11:00 a.m.
Morning tea
11:00 a.m. – 11:45 a.m.
Stimulating biomolecular systems II
Prof Alan Mark
11:45 a.m. – 12:30 p.m.
Computing and software resources for molecular simulations Prof Bernard Pailthorpe
12:30 p.m. – 02:00 p.m.
Lunch
02:00 p.m. – 03:00 p.m.
Effective use of the RCSB Protein Data Bank
Prof Phil Bourne
03:00 p.m. – 04:00 p.m.
Discussion session
Thursday 28 June 2007
Statistical analysis of gene expression
09:00 a.m. – 10:00 a.m. From gene expression to clinical diagnostic tool - are
we there yet?
Prof Sue Wilson
10:00 a.m. – 10:30 a.m.
Morning tea
10:30 a.m. – 11:15 a.m. Borrowing strength in microarray data analysis
Dr Gordon Smyth
11:15 a.m. – 12:00 p.m. Detection of differential expression with microarray
data
Prof Geoff McLachlan
12:00 p.m. – 1:30 p.m.
Lunch
01:30 p.m. – 02:15 p.m.
A myogenin network-centric systems biology approach to the genetic dissection of complex traits in beef cattle
Dr Antonio Reverter
TIMETABLE
3 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
02:15 p.m. – 03:00 p.m. “Differential expression free” analysis of microarray
data
Dr Harri Kiiveri
03:00 p.m. – 03:30 p.m.
Afternoon tea
03:30 p.m. – 04:15 p.m.
Assessing classifiers trained on gene expression data
Dr Ian Wood
05:00 p.m. – 06:30 p.m.
Plenary talk – Ligand binding site searching and application to finding off-targets for major pharmaceuticals
Prof Phil Bourne
06:30 p.m. – 07:30 p.m.
Refreshments
Friday 29 June 2007
Computational neuroscience
09:00 a.m. – 10:15 a.m.
Is your brain smarter than a computer? Introduction to neuroscience and computational neuroscience
A/Prof Geoff Goodhill
10:15 a.m. – 10:45 a.m.
Morning tea
10:45 a.m. – 11:45 a.m. Optimization principles of adaptive coding in the
primary visual cortex
Dr Tatyana Sharpee
11:45 p.m. – 01:15 p.m.
Lunch
01:15 p.m. – 02:15 p.m.
Smart computations in small brains: vision, navigation, perception and cognition in honeybees
Prof Mandyam Srinivasan
02:15 p.m. – 03:15 p.m.
Towards a theory of learning and levels for neurobiology
Dr Anthony Bell
03:15 p.m. – 03:30 p.m.
Afternoon tea
03:30 p.m. – 04:30 p.m.
Network structure of cerebral cortex shapes neuronal dynamics on multiple time scales
A/Prof Michael Breakspear
~*~*~*~*~*~
ABSTRACTS Monday 25 June 2007 Bio-image analysis, quantification and classification
4 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
09:40 a.m. – 10:10 a.m.
Speaker:
Dr Rohan Teasdale ARC Centre in Bioinformatics, and Institute for Molecular Bioscience The University of Queensland
Bio:
Rohan Teasdale heads the "Computational Cell Biology" research group at Institute for Molecular Bioscience, University of Queensland. One of the innovations of his research group is a synergetic approach that combines experimental cell biology and microscopy with computational methods. His research group is focused on understanding how individual proteins are compartmentalised and defining the protein machinery responsible for their transport. He recently developed the LOCATE database (http://locate.imb.uq.edu.au/), a resource focused on all aspects relevant to understanding a proteins subcellular localisation.
Title:
An introduction to cellular imaging techniques
Abstract:
I will introduce the types of imaging techniques currently in use within the bio-medical field. Illustrative examples of research applications based on my group’s cellular imaging will be presented. Topics covered will include (1) basics of fluorescence; (2) microscopy types; (3) labelling techniques, and (4) selective applications. Recommended online resources include: http://www.olympusmicro.com/index.html http://www.microscopyu.com/ http://probes.invitrogen.com/resources/education/
ABSTRACTS Monday 25 June 2007 Bio-image analysis, quantification and classification
5 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
10:10 a.m. – 11:40 a.m.
Speaker:
Dr Estelle Glory Centre for Bioimage Informatics Carnegie Mellon University, Pittsburgh, PA, USA
Bio:
Estelle Glory is a postdoctoral fellow in the Center for Bioimage Informatics in Carnegie Mellon University. She joined the group of Robert F. Murphy after receiving her PhD in biological image analysis, working with J-C. Olivo-Marin (Institut Pasteur, Paris, France) and G. Stamon (University Paris 5). With a background in biochemistry and computer science, her work is focused on the extraction of quantitative information from microscopy images to analyze and interpret high throughput experiments. Her current research is related to the determination of subcellular location patterns using cell segmentation, feature extraction and machine learning methods.
Title:
Automated interpretation of subcellular patterns in microscope images: bioimage informatics for systems biology
Abstract:
In systems biology, protein structures, protein interaction, expression level have been largely analyzed for understanding and modeling cell pathways while protein location, which is critical information to build such spatiotemporal models has been less explored. The prediction of protein location from sequences is limited by the samples of the training set while the human annotations are restricted by the predefined vocabulary, for example the Gene Ontology. The Murphy lab has pioneered the application of machine learning methods to protein images. The automatic extraction of features from fluorescence microscope images has been developed to provide objective, accurate and reproducible description of protein patterns. The features have been designed to be robust enough to non informative characteristics, such as the rotation or the absolute position of protein, but sensitive enough to small variations characterizing slightly different patterns. These quantitative descriptors are used as input to different tools developed to automatically analyze high throughput experiments. Particularly useful tools include statistical analysis of the difference between two patterns, supervised classification to recognize subcellular compartments and the clustering of proteins into families that share the same protein pattern. The next step is the creation of models which capture the essence and variation of protein location patterns. This is achieved with a generative model. Beyond the compactness of the structured information, such a model is able to generate new cell images, combining the nuclear shape and texture, the cell boundary, and the relative location and density of proteins. The possibility to generate artificial cells containing thousands of protein patterns will doubtless be a major contribution in the field of systems biology. The goal of bioimage informatics is to provide a direct path for generating biological knowledge from images with a minimum of human intervention, and significant progress has been made towards this goal.
ABSTRACTS Monday 25 June 2007 Bio-image analysis, quantification and classification
6 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
01:15 p.m. – 02:15 p.m.
Speaker:
A/Prof Tuan Pham Director, Bioinformatics Applications Research Centre James Cook University
Bio:
Tuan D. Pham is an Associate Professor in the School of Mathematics, Physics, and Information Technology; and Director of the Bioinformatics Applications Research Centre at James Cook University. His research experience and interests are diverse which cover image processing, pattern recognition, signal processing, fuzzy logic, neural networks, genetic algorithms, and applied geostatistics with applications to bioinformatics and biomedical informatics. He has contributed pioneering research work on fuzzy finite element analysis of engineering problems; and applications of predictive coding and geostatistical models for analysis of bioimaging, microarray gene-expression, and mass-spectrometry data.
Title:
On models and algorithms for analysis of biological images
Abstract:
Recent advances in modern biotechnology offer interesting and challenging problems to computational scientists with respect to the handling and interpretation of complex biological data. Solutions to these problems are anticipated to revolutionize our way of living in the sense that human fatal diseases can be early detected and diagnosed for proper treatments, new therapeutic drugs be discovered and personalized medicine be developed. This talk will particularly address some recently developed computational models and algorithms for bio-imaging analysis and classification. It will highlight the importance of the incorporation of the skills and knowledge gained from biology, biomedicine, mathematics, engineering, computer and information sciences.
ABSTRACTS Monday 25 June 2007 Bio-image analysis, quantification and classification
7 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
02:15 p.m. – 03:15 p.m.
Speaker:
Ms Leanne Bischof Image Analyst, Biotech Imaging CSIRO Mathematical and Information Sciences
Bio:
Leanne Bischof is a senior member of the Biotech Imaging group at CSIRO Mathematical and Information Sciences. The group conducts image analysis research into segmentation and feature extraction techniques and develops commercial software for biotechnology applications, in particular for High Content Analysis (HCA). HCA refers to the high throughput analysis of fluorescence microscope images of cells to quantify cellular morphology and function. There are a handful of companies which supply HCA systems to the pharmaceutical industry to screen candidate drugs for efficacy and toxicity. The group's HCA software has been licensed to several of these companies.
Title:
High content cellular screening for drug discovery
Abstract:
In modern biology, important biological information is often captured in the form of images. Extracting the information from those images manually can often be tedious and time consuming. There is increasing demand for software to perform this analysis automatically. Modern image analysis techniques are making it possible to automate even the most challenging applications. We will illustrate what is possible by referring to our work in High Content Analysis (HCA). HCA refers to the automated analysis of mainly fluorescence microscope images of cells to quantify cellular morphology and function. High content analysis is used in the pharmaceutical industry to screen candidate drugs for efficacy and toxicity. It is increasingly being used in academia to expand the fundamental understanding of cellular biology. We will briefly mention some of the image segmentation and feature extraction techniques that we use and then show a series of biological assays which require a range of analysis techniques. These HCA assays will include neurite outgrowth analysis in 2D and 3D, analysis of mixed cell populations (such as neuron-astrocyte co-cultures or differentiating neural stem cells), analysis of protein translocation, co-localisation and sub-cellular localisation, and tracking of proteins over time (such as vesicles in TIRF microscope images). We will briefly canvas the software engineering challenges inherent in developing image analysis software for a range of environments - for direct use by biologists (such as our standalone HCA-Vision package), for integration into commercial HCA systems (where the host system provides the user interface) and for our in-house use to support one-off solutions for our research collaborators.
ABSTRACTS Monday 25 June 2007 Bio-image analysis, quantification and classification
8 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
03:45 p.m. – 04:45 p.m.
Speakers:
Mr Oliver Cairncross Mr Andrew Noske ARC Centre in Bioinformatics, and Institute for Molecular Bioscience The University of Queensland
Bio:
The Visible Cell™ project based at the Institute for Molecular Bioscience (IMB) and the ARC Centre in Bioinformatics (ACB) at the University of Queensland (UQ) represents a large-scale, cross-disciplinary, multi-institutional and international e-Science initiative that is changing the way we think about mammalian cells. The Visible Cell™ project aims to inform advanced in silico studies of cell and molecular organisation in 3D using the mammalian cell as a unitary example of an ordered complex system. This unique initiative is founded on the provision of complete sets of 3D spatio-temporal coordinates for whole mammalian cells at a range of resolutions and the integration of data on gene products, molecular interactions, pathways, networks and processes into the corresponding cellular coordinates. Investigators will interact with the cellular structures, molecules and processes (driven by user-supplied computational models) inside an integrated visualisation environment.
Title:
Reconstructing the mammalian cell
Abstract:
The Visible Cell™ project based at the Institute for Molecular Bioscience (IMB) and the ARC Centre in Bioinformatics (ACB) at the University of Queensland (UQ) represents a large-scale, cross-disciplinary, multi-institutional and international e-Science initiative that is changing the way we think about mammalian cells. The Visible Cell™ project aims to inform advanced in silico studies of cell and molecular organisation in 3D using the mammalian cell as a unitary example of an ordered complex system. This unique initiative is founded on the provision of complete sets of 3D spatio-temporal coordinates for whole mammalian cells at a range of resolutions and the integration of data on gene products, molecular interactions, pathways, networks and processes into the corresponding cellular coordinates. Investigators will interact with the cellular structures, molecules and processes (driven by user-supplied computational models) inside an integrated visualisation environment. In this talk the foundations of the Visible Cell™ will be outlined and a software demonstration given.
ABSTRACTS Tuesday 26 June 2007 Modelling and simulation of cellular processes
9 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
09:00 a.m. – 10:00 a.m.
Speaker:
Prof Kevin Burrage ARC Federation Fellow Advanced Computational Modelling Centre ARC Centre in Bioinformatics, and Institute for Molecular Bioscience The University of Queensland
Bio:
Kevin Burrage is a Federation Fellow of the Australian Research Council. He has joint positions within Mathematics and the IMB at the University of Queensland, is the Director of the Advanced Computational Modelling Centre and is one of the CIs in the ARC Centre of Excellence in Bioinformatics. His main research interests are in Computational and Systems Biology and Computational Science in general and has a specific interest in the role of noise in cellular dynamics. He has over 160 published scientific articles.
Title:
The role of noise in cellular processes
Abstract:
This talk will give a brief introduction to the sorts of noise processes that arise in Cell Biology. We will discuss the nature of these noise processes and how they can be modelled and simulated. We will focus on models from genetic regulation and cascading reactions in the cytosol in both discrete and continuous, and delayed and non-delayed setting. No prior knowledge of stochastic modelling is needed and the talk will focus on concepts rather than mathematical intricacies.
ABSTRACTS Tuesday 26 June 2007 Modelling and simulation of cellular processes
10 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
10:30 a.m. – 11:30 p.m.
Speaker:
Dr Edmund Crampin Bioengineering Institute and Department of Engineering University of Auckland, New Zealand
Bio:
Edmund Crampin is Senior Lecturer at the Auckland Bioengineering Institute where he leads the Systems Biology and Cell Modelling research group. His current research interests include mathematical modelling of metabolic, signalling and genetic networks, with a particular focus on computational modelling of cardiac myocytes. Edmund completed a DPhil in Applied Mathematics at Oxford. He was a Junior Research Fellow of Brasenose College, Oxford and was awarded a Research Fellowship from the Wellcome Trust to study mathematical models of heart disease. Edmund joined the University of Auckland as a Research Fellow and was subsequently appointed to a lectureship jointly between the Department of Engineering Science and the Auckland Bioengineering Institute.
Title:
Virtual heart disease
Abstract:
The heart is a relatively simple organ. It is composed predominantly of a single type of cell – the cardiac myocyte – and it performs a single function – to pump blood around the circulation. The mechanism by which electrical stimulation of a myocyte leads to generation of the action potential and, subsequently, to contraction of the muscle cell, is relatively well understood, and is well described in mathematical models. Reduction of the usual supply of oxygenated blood to a region of the heart muscle can have a devastating effect on the heart's ability to pump. The complex sequence of events in ischaemic heart disease, arising from the obstruction of a coronary artery and leading to life-threatening pump failure, are much less well known, despite the wealth of available experimental data. In this talk I will discuss our approach to mathematical modelling of the pathophysiology of cardiac myocytes during ischaemia, by focusing on key events including the acidification of the tissue (acidosis) and build-up of extracellular potassium (hyperkalaemia) that occur when the blood supply is reduced. I will show how these and other aspects of myocyte dysfunction in ischaemic heart disease can be understood through biophysically-based computational modelling of heart cells, and discuss how this may ultimately lead to a better understanding of the progression of ischaemic heart disease.
ABSTRACTS Tuesday 26 June 2007 Modelling and simulation of cellular processes
11 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
01:30 p.m. – 02:30 p.m.
Speaker:
Dr Markus Hegland Centre for Mathematics and its Applications Mathematical Sciences Institute The Australian National University
Bio:
Markus Hegland is a computational mathematician working at the ANU. His main interests relate to high-dimensional, in particular sparse grid approximation, with applications in computational biology and machine learning. He has also worked in the areas of ill-posed and inverse problems and high performance computing.
Title:
The numerical solution of the master equations in molecular biology – how, why and for what
Abstract:
It is now widely accepted that molecular processes in biology are a fundamentally noisy affair and are thus best modelled by stochastic processes. Here, we consider random fluctuations in protein and RNA levels. Interesting biological questions relate to how these levels change over time – in particular in response to external and internal stimuli -- and how these levels ultimately settle in on some stationary values. Both the state of the system at a fixed time and the time it takes to arrive at a stationary state are random variables and are characterised by an expected value or mean and a scale parameter like the variance or quantiles. These features of the stochastic process are traditionally obtained by simulation. It turns out that for complex systems the determination of these features is computationally demanding due to high numbers of replicated simulations of many interacting processes. Here we consider an alternative to simulation -- the numerical approximation of evolving probability distributions characterising the stochastic processes. The governing equations for the probability distributions are the chemical master equations. Often, the curse of dimensionality is cited as the main obstacle to the approximation of probability distributions and to the solution of the chemical master equations. Here, we will discuss methods to numerically approximate and solve these equations. In particular, we will introduce a method based on combining simple approximations which is capable to handle the curse of dimensionality to some extent. We will show that this approach is feasible for the solution of simple master equations involving 100 different substances. At the end of this presentation, the student should have some idea about biological questions which can be answered with this approach, the inherent computational challenges and the tools used to address these challenges.
ABSTRACTS Tuesday 26 June 2007 Modelling and simulation of cellular processes
12 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
03:00 p.m. – 03:45 p.m.
Speaker:
Dr André Leier Advanced Computational Modelling Centre The University of Queensland
Bio:
André Leier is a postdoctoral fellow in the Advanced Computational Modelling Centre at the University of Queensland, with research interests in Computational and Systems Biology, in particular stochastic, spatiotemporal, and multi-scale models of cell signalling and genetic regulation and the roles of delay in cellular processes. He studied Computer Science and Mathematics and received his PhD in Computer Science from the University of Dortmund, Germany.
Title:
Stochastic modeling and simulation of cellular processes with delays
Abstract:
Time delays associated with slow biochemical processes such as transcription, translation, nuclear and cytoplasmic translocations are known to affect the dynamics of genetic regulation. Temporal models of cellular signalling and genetic regulation have to take these delays into account in order to capture the dynamics more accurately and to allow for more reliable predictions. In this talk, we discuss stochastic delay modelling and simulation suitable to capture both stochasticity and delay in temporal models of cellular signalling and genetic regulation.
ABSTRACTS Tuesday 26 June 2007 Modelling and simulation of cellular processes
13 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
03:45 p.m. – 04:30 p.m.
Speaker:
Mr Paul Taylor The MathWorks Australia Pty Ltd
Bio:
Paul Taylor is a senior applications engineer at MathWorks Australia. MathWorks is the leading global provider of software for technical computing and model-based design. Paul will present a workshop on MATLAB, a powerful interactive matrix-based environment for scientific and engineering modelling and computation. MATLAB has a large user base in Australia and overseas, and extensions are available for a number of application domains including bioinformatics.
Title:
Modeling and simulation in computational biology
Abstract:
Biological data has become so diversified and complex that flexible, non-niche data analysis and visualization tools are critical to the success of most biological research. The MathWorks’ products for computational biology provide a user-friendly, flexible programming platform for analyzing complex biological data and systems. From a single environment a broad range of analysis, simulation, algorithm development can be undertaken to radically accelerate the research and discovery process. This presentation will highlight the benefits of SimBiology, a flexible environment for modelling, simulating and analysing biochemical pathways, and the MATLAB Distributed Computing Engine, enabling users to easily distribute large simulations across a computer cluster to dramatically reduce computation time.
ABSTRACTS Wednesday 27June 2007 Prediction and modelling of protein structure and dynamics
14 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Sessions:
09:00 a.m. – 09:45 a.m. 11:00 a.m. – 11:45 a.m.
Speaker:
Prof Alan Mark ARC Federation Fellow Centre in Computational Molecular Sciences School of Molecular and Microbial Sciences and Institute for Molecular Bioscience The University of Queensland
Bio:
Prof. Alan E. Mark’s primary interest is in understanding how biological systems are regulated at an atomic level. He studied Chemistry and Biochemistry at the University of Sydney before undertaking a PhD in Physical Biochemistry at the ANU. After postdoc's at ANU, Groningen, The Netherlands and ETH, Zurich Switzerland he was appointed to a chair of Biomolecu;ar Simulation at the University of Groningen. In 2005 he moved to UQ on an ARC Federation Fellowship. The main focus of his current research is in understanding how biomolecular systems self-organize, in particular he uses the atomistic simulations to investigate processes such a protein and peptide folding, lipid aggregation, protein-ligand interactions and signal transduction processes within cells. Prof. Mark is closely associated with the GROMOS and GROMACS molecular simulation packages and the development of the GROMOS empirical force field.
Title:
Stimulating biomolecular systems
Abstract:
The lectures will provide a general introduction to the simulation of the dynamics of protein and lipid systems focusing in particular on how biomolecular systems are best represented in atomic or near atomic detail and the types of motion can be observed on currently accessible time scales. Calling on a wide range of example the lectures will illustrate what types of information can be obtained using atomistic simulation techniques and demonstrate how such simulations are not only enabling a more detailed interpretation of experimental data but in some cases also challenging our basic assumptions regarding how biomolecular systems work. Of equal importance the lectures will show examples that highlight dangers and pit falls of using simulations inappropriately.
ABSTRACTS Wednesday 27June 2007 Prediction and modelling of protein structure and dynamics
15 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
09:45 a.m. – 10:30 a.m.
Speaker:
Dr Thomas Huber Computational Biology and Bioinformatics Environment School of Molecular and Microbial Sciences The University of Queensland
Bio:
After his PhD studies, which were partly conducted at the Australian National University, Dr Huber returned to the ETH Zurich for a one year post-doctoral stint before in 1998 taking up a permanent position as computational chemist at the Australian National University’s Supercomputing Facility. In 2001 Dr Huber joint the University of Queensland, taking up a lectureship in computational biology / bioinformatics The main theme of his group’s work is centred in protein structure prediction. Protein structure predictions are still quite limited and additional (sparse) distance constraints generally greatly improves the quality of predictions. Currently they explore several sources of swiftly to obtain, additional information to aid predictions: Residue contact information from sequence evolution, distance information from chemical crosslinking, and structural restraints from paramagnetic NMR spectroscopy. In the framework of UQ’s Structural Genomics Programme, they work on identifying protein properties that affect high-throughput structure determination, and use these insights predictively to minimize protein loss due to insolubility and to maximize success in protein expression and crystallization. Recently, they also branched into the exciting new area of computational microbial ecology where we started to computationally analyze meta-genomics shotgun sequencing data from whole microbial communities.
Title:
Determination of protein complexes and multi-domain proteins using a combination of experiments and calculations
Abstract:
In biology the concept of bottom-up integration has inspired large scale programmes to make inventories of all molecular components in a cell, most publicized whole genome sequencing projects. More recently, numerous large scale initiatives have been launched to determine protein structures in high throughput, but while these initiatives have highly accelerated the structure determination process, such programmes are also heavily biased towards determining the structures of small, soluble, easy to crystallise and single chain proteins. To this end, very little insight is gained on how a protein interacts with other proteins, an information which in many cases is crucial to understand biological function. New biochemical experiments, most notable the two-hybrid methods and bait-tag purification approaches, have been developed to answer this fundamental question which proteins physically interact, and while they are very powerful to produce binary (yes/no) answers on large sets of proteins they fail to explain how proteins interact at an atomic level. The focuses of this talk will be on approaches to combine limited experimental data with molecular modelling to determine structure of protein complexes and multi-domain proteins at atomic level. I will introduce and review these methods and illustrate results with calculations on real proteins. Finally, limitations are outlined by following the time honoured tradition of trying to understand things by breaking them.
ABSTRACTS Wednesday 27June 2007 Prediction and modelling of protein structure and dynamics
16 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
11:45 a.m. – 12:30 p.m.
Speaker:
Prof Bernard Pailthorpe CEO, Queensland Cyber Infrastructure Foundation Chair of Computational Science ARC Centre in Bioinformatics The University of Queensland
Bio:
Prof. Bernard Pailthorpe is a physicist who founded Sydney VisLab in 1992, to support computational and visualisation research across a broad spectrum of disciplines. During 1999-2000 he directed the scientific visualisation program for NPACI and SDSC at UCSD (USA). The group efforts there were focused on scalable volume visualization, and participated in the opening show for the Hayden Planetarium in New York. He now holds the Foundation Chair of Computational Science at UQ and is CEO of QCIF Ltd, an Australian HPC Consortium that supports industry and research projects, and develops cyberinfrastructure. He has wide experience in physics education and developing new classes in Computational Physics. He has advised Governments at senior levels on HPC and e-Research, leading to a new funding program to establish the Australian Partnership for Advanced Computing in 2000.
Title:
Computing and software resources for molecular simulations
Abstract:
Molecular simulations in diverse field (bioscience, biomedicine, chemistry and others) require advanced computing, networking and data management, as well as community development and support. This presentation will provide an update on all these aspects of computing and software resources in Australia, particularly those being developed and organised under the National Collaborative Research Infrastructure Scheme.
ABSTRACTS Wednesday 27June 2007 Prediction and modelling of protein structure and dynamics
17 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
02:00 p.m. – 03:00 p.m.
Speaker:
Prof Phil Bourne Co-Director, Protein Data Bank (PDB) Department of Pharmacology, and San Diego Supercomputer Center University of California, San Diego, USA
Bio:
Philip E. Bourne PhD is a Professor in the Department of Pharmacology at the University of California San Diego, Co-director of the Protein Data Bank and an Adjunct Professor at the Burnham Institute and the Keck Graduate Institute. He is a Past President of the International Society for Computational Biology. He is an elected fellow of the American Medical Informatics Association. He is the Founding Editor-in-Chief of the open access journal PLoS Computational Biology, on the Advisory Board of Biopolymers and on the Editorial Boards of Proteins: Structure Function and Bioinformatics, Biosilico and IEEE Trends in Computational Biology and Bioinformatics and a long standing member of the National Science Foundation and National Institutes of Health panels responsible for reviewing proposals relating to biological infrastructure and bioinformatics. He is a past member of the US National Committee for Crystallography, past chairman of the International Union of Crystallography Computing Commission IUCrCC and past chairman of the American Crystallography Association (ACA) Computing Committee. Recent awards include the Flinders University Convocation Medal for Outstanding Achievement 2004 and the Sun Microsystems Convergence Award 2002. Bourne's professional interests focus on bioinformatics and structural bioinformatics in particular. This implies algorithms, metalanguages, biological databases, biological query languages and visualization with special interest in evolution, cell signaling and apoptosis. He has published over 180 papers and 4 books, one of which sold over 120,000 copies. He has co-founded 3 companies: ViSoft Inc., Protein Vision Inc. and a company distributing independent films for free.
Title:
Effective use of the RCSB Protein Data Bank (PDB)
Abstract:
The RCSB PDB Web site at http://www.pdb.org provides access to all publically accessible data on the structure of biological macromolecules. To make these data most useful they have been integrated with at least 30 other sources of information [1]. Most recently the complete PDB dataset has been remediated to provide correspondence to the UniProt protein sequence, consistent features of ligands, standard nomenclature etc. I will provide a tour of these resources by way of addressing example questions at different levels of complexity. [1] N. Deshpande, et al. 2005 The RCSB Protein Data Bank: A Redesigned Query System and Relational Database Based on the mmCIF Schema Nucleic Acids Research. 33: D233-D237.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
18 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
09:00 a.m. – 10:00 a.m.
Speaker:
Prof Sue Wilson Co-Director, Centre for Bioinformation Science The Australian National University
Bio:
Susan Wilson is Director of the Centre for Bioinformation Science, Mathematical Sciences Institute, The Australian National University (ANU). She obtained her B.Sc. from the University of Sydney, followed by her Ph.D. from the ANU in 1972, and then spent two years as a Lecturer in the Department of Probability and Statistics at Sheffield University. She returned to ANU towards the end of 1974 and has since held a variety of positions there, both in some of the Statistical Science groupings, as well as at the National Centre for Epidemiology and Population Health, and now heads the bioinformatics research facility at ANU. Sue has over 150 refereed publications in biometry and applied statistics, with a particular emphasis in statistical genetics/genomics. Many papers have arisen from her extensive consulting experience in the biological, social and medical sciences, leading to statistical modelling developments to answer substantive research questions in these disciplines.
Title:
From gene expression to clinical diagnostic tool - are we there yet?
Abstract:
The new 'omic' technologies have the potential to analyse your genome and cell processes so that decisions can be made as to whether to treat you, and if so how to choose the best course of treatment. The medical aim is to optimise your potential positive outcome/s while minimising any adverse effects. The resultant era in the offing is being called "personalised medicine". To underpin this medical progress, high quality statistical approaches need to be applied to the data on which these developments are based. The statistical evidence underpinning one of the first drugs licensed by the U.S. Food and Drug Administration (FDA) as a 'significant step towards personalised prescribing' has been judged as seriously flawed (Ellison in "Significance", 2006). More recently the FDA approved a cancer prognosis test based on gene expression microarray technology - a first. The claim is that the prognostic profile so produced potentially provides a powerful tool to tailor adjuvant systemic treatment that could greatly reduce the cost of (breast) cancer treatment, both in terms of adverse side effects and health care expenditure. So, what is the statistical evidence? Following a brief overview of microarray technology, including the fundamental importance of the need for careful quality control, the statistical evidence and challenges for development of predictive gene lists, often termed 'gene signatures' will be reviewed. Particular emphasis will be placed on how the signature produced depends on the selection of patients in the 'training set', and the lack of agreement when comparing the signature lists from different studies.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
19 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
10:30 a.m. – 11:15 a.m.
Speaker:
Dr Gordon Smyth Senior Research Scientist, Bioinformatics Division Walter and Eliza Institute for Medical Research
Bio:
Gordon Smyth is a Senior Research Fellow and Lab Head at the Walter and Eliza Hall Institute of Medical Research. He develops statistical methods for microarray data analysis and functional genomics. He is the author of the popular limma software package for R for linear modelling of microarray data.
Title:
Borrowing strength in microarray data analysis
Abstract:
At a molecular biology research institute, most microarray experiments are differential expression studies. Such studies can range from simple in design, perhaps comparing just two groups, to more complex designs involving multiple levels of several treatments. Whether simple or complex, these experiments invariably involve only a small number of biological replicates. This means that creative ways to borrow strength between genes and between samples are essential to the statistical analysis. This talk will describe an approach which has proved popular and effective, using linear models to borrow strength between samples, and empirical Bayes methods to borrow strength between genes. This approach leads to an elegant generalization of t-tests and F-tests. The talk will go on to consider information borrowing ideas for experiments with multiple error strata, and gene set tests which conduct hypothesis tests for ensembles of genes.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
20 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
11:15 a.m. – 12:00 p.m.
Speaker:
Prof Geoff McLachlan ARC Centre in Bioinformatics, and Department of Mathematics The University of Queensland
Bio:
Geoff McLachlan is Professor of Statistics in the Department of Mathematics and a Professorial Research Fellow in the Institute for Molecular Bioscience. He is also a chief investigator in the ARC Centre of Excellence in Bioinformatics. His research has been recognized with various awards, including a DSc in 1994 and an ARC Professorial Fellowship in 2006. He has written over 165 articles, including 5 monographs, the last four as volumes in the prestigious Wiley series in Statistics. His research in statistics has been concentrated on the related fields of classification and machine learning, and in the field of statistical inference. The focus in the latter field has been on the theory and applications of finite mixture models and on estimation via the EM algorithm. More recently, he has become actively involved in the field of bioinformatics with the focus on the statistical analysis of microarray gene expression data in which he has coauthored a Wiley monograph.
Title:
Large-scale simultaneous inference for the detection of differential expression with microarray data
Abstract:
Microarrays allow the measurement of gene expressions for a biological sample (tissue) on a genome-wide scale, and form part of the high-throughput -omics methodology which is changing the face of biological research (genomics, proteomics and metabonomics). They are now standard tools in biology, with an ultimate goal for their use in clinical medicine for diagnosis and prognosis, in particular in cancer towards guiding therapeutic management. In this talk we consider an important problem in microarray experiments concerning the detection of genes that are differentially expressed in a given number of classes. It requires large-scale hypothesis testing problems, with hundreds or thousands of test statistics to consider at once. We consider the use of normal mixture models that provide a framework for the correct choice of a null distribution for simultaneous significance testing, and its effect on inference. This methodology is demonstrated on some real microarray data sets published in the literature.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
21 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
01:30 p.m. – 02:15 p.m.
Speaker:
Dr Antonio Reverter Principal Research Scientist, Bioinformatics Group CSIRO Livestock Industries
Bio:
Dr Antonio (Toni) Reverter is a principal research scientist with the Bioinformatics Group of CSIRO Livestock Industries. Toni's background is in statistical genetics, more specifically in methods for large-scale genetic evaluation and parameter estimation. His work in CSIRO involves the development and application of mathematical, computational and statistical methods for the analysis of gene-expression and mapping data including whole-genome SNP genotypes for complex traits in livestock species. Toni was the recipient of the inaugural 2005 Eureka Prize for Bioinformatics.
Title:
A myogenin network-centric systems biology approach to the genetic dissection of complex traits in beef cattle
Abstract:
Despite the advances that have rendered new genetic technologies attractive to many animal geneticists it is still unclear how to best analyse the resulting data productively. The futility of simple statistical abstraction of genetic effects and sole reliance of genetic association on statistical significance alone are now apparent. This disappointing outcome is particularly obvious when dealing with complex traits governed by many interacting genetic effects. In order to shift genetic improvement of livestock species to improved frameworks, it is imperative to incorporate sound knowledge of the biology at the molecular genetics level of the traits of economic importance. Motivated by the added value of genetical genomics studies that merge expression profiling with marker-based genotyping, we propose a systems biology approach anchored to a gene network for Myogenin (MYOG), a muscle-specific transcription factor essential for the development of skeletal muscle. Using bovine gene expression and high density marker data, our objective is to evaluate the strength of the relationship between the association of a single nucleotide polymorphism (SNP) to a phenotype of interestwith the transcriptional activity of genes in the network.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
22 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
02:15 p.m. – 03:00 p.m.
Speaker:
Dr Harri Kiiveri Statistical Bioinformatics - Health CSIRO Mathematical and Information Sciences, Floreat, Western Australia
Bio:
Dr Harri Kiiveri is a research statistician who develops statistical methods for analysing very high dimensional multivariate data. He currently works in the area of bioinformatics with CSIRO Mathematical & Information Sciences (CMIS), specifically with the analysis of microarray data and SNP data. He has developed a methodology for fitting a large class of statistical models to data sets with many more variables than observations. He has also developed methods for the construction of local and global gene networks which enable the integration of different data sources such as microarrays, protein - protein interactions and transcription factor binding site information.
Title:
“Differential expression free” analysis of microarray data
Abstract:
In this talk we'll consider a way of analysing microarray data which does not focus on differential expression. Guided by a response of interest, we'll look at fitting models to data sets with many more variables than observations, constructing local gene networks, and simulations from such networks as a means of understanding the data and generating new hypotheses.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
23 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
03:30 p.m. – 04:15 p.m.
Speaker:
Dr Ian Wood Research Fellow in Mathematical Sciences Queensland University of Technology
Bio:
Ian Wood is a researcher based in the ARC Center in Bioinformatics at the University of Queensland. He completed a PhD in 2004 at the University of Queensland on Boltzmann machines, optimisation and Markov chain Monte Carlo methods. He then completed a three-year postdoctoral fellowship in statistical genetics at Queensland University of Technology, which included collaboration with Genetic Solutions Pty Ltd and QIMR. His research interests include analysis of gene expression data, comparative genomics, classification, Monte Carlo methods, mixture models and machine learning.
Title:
Assessing classifiers trained on gene expression data
Abstract:
Levels of gene expression are typically estimated through the proxy of mRNA levels as measured after hybridization in microarray experiments. When measured on a set of subjects, they produce datasets which usually contain few observations (n<100) and thousands of possible predictors (p>>n). Studies have used microarray data for problems such as diagnosis and prognosis of disease and classification of tumours into subtypes. In constructing a classifier, the goal is to minimise and assess the error rate expected on new data. Methods of assessment such as cross-validation split the data into training and test sets. The test data should not be used in the choice of any aspect of the classifier being assessed. Failure to do this leads to selection biases of varying severity. Methods to detect and avoid these will be described.
ABSTRACTS Thursday 28 June 2007 Statistical analysis of gene expression
24 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
05:00 p.m. – 06:30 p.m. – Plenary talk
Speaker:
Prof Phil Bourne Co-Director, Protein Data Bank (PDB) Department of Pharmacology, and San Diego Supercomputer Center University of California, San Diego, USA
Bio:
Refer to page 17.
Title:
Ligand binding site searching and application to finding off-targets for major pharmaceuticals
Abstract:
We have recently developed a new method we refer to as the geometric potential for defining ligand binding sites in cases where a ligand-receptor complex exists [1]. Using this descriptor and a sequence order independent profile-profile analysis (SOIPPA) approach we have been able to uncover new evolutionary relationships between families of proteins. [2]. Since a variety of major pharmaceuticals are found in the PDB bound to receptors, in addition we have an excellent approach for studying off-site binding of these ligands. When coupled with biochemical evidence and clinical evidence of side effects interesting stories emerge. I will describe the tamoxifen and other stories [3]. [1] L. Xie and P.E.Bourne 2007 A robust and efficient algorithm for the shape description of
protein structures and its application in predicting ligand binding sites. BMC Bioinformatics, 8(Suppl 4):S9
[2] L. Xie and P.E. Bourne 2007 Detecting Evolutionary Linkages Across Fold and Functional
Space with Sequence Order Independent Profile-profile Alignments. Submitted. [3] L. Xie and PE. Bourne 2007 Proteome-wide Elucidation of the Molecular Mechanism
Defining the Adverse Effect of Selective Estrogen Receptor Modulators. Submitted.
ABSTRACTS Friday 29 June 2007 Computational neuroscience
25 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
09:00 a.m. – 10:15 a.m.
Speaker:
A/Prof Geoff Goodhill Queensland Brain Institute The University of Queensland
Bio:
Geoff Goodhill is an Associate Professor in both the Queensland Brain Institute and the School of Physical Sciences at the University of Queensland. His research focuses on understanding the computational mechanisms controlling the development of connections between neurons. He is particularly interested in how the tips of growing nerve fibres are guided by molecular gradients in the developing nervous system, and how the statistics of the visual input influence the structure of topographic maps in the visual cortex.
Title:
Is your brain smarter than a computer? Introduction to neuroscience and computational neuroscience.
Abstract:
Biological nervous systems are far more powerful, robust and efficient computing devices than any computers humans have yet designed. The goal of computational neuroscience is to understand the nature of the computations nervous systems have to perform, how they perform them, and how this can lead to better silicon-based computers. In my talk I will introduce some of the amazing computational abilities of biological brains, and some of the main themes in current computational neuroscience research. Illustrations of these issues will include recent results from my lab on the computational principles underlying the development of neuronal wiring.
ABSTRACTS Friday 29 June 2007 Computational neuroscience
26 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
10:45 a.m. – 11:45 a.m.
Speaker:
Dr Tatyana Sharpee Salk Institute for Biological Studies San Diego, CA, USA
Bio:
Tatyana Sharpee is an assistant professor in the Computational Neurobiology laboratory at the Salk Institute for Biological Studies. Her research focuses on statistical physics and information theory approaches to characterizing signal processing in the nervous system. In particular, she is interested in how sensory processing in the brain is matched to the statistics of real-world signals, why might the evolved hierarchy of neural representations be optimal, and how it can be best adapted to track rapid changes in the statistics of inputs. She applied information-theoretic methods to characterize neural feature selectivity with natural stimuli, and showed that certain aspects of neural filtering in visual cortex could adapt to even fairly complex statistical parameters in natural scenes.
Title:
Optimization principles of adaptive coding in the primary visual cortex
Abstract:
The idea that adaptation serves to adjust properties of neurons and their populations to optimally encode incoming stimuli is one of the central and oldest in sensory neuroscience, dating back to Adrian. However, most theoretical predictions for specific parameters of neural sensitivity, such as receptive fields, were made using a linear (or threshold-linear) model of neural response. While these predictions adequately describe sensitivity of sensory neurons at the periphery (e.g. retina and lateral geniculate nucleus), we have found that cortical neurons exhibit adaptive filtering with qualitative new features that do not fit with optimal encoding arguments for a linear system. Instead, we have assumed that neural response can be described by a generalized linear-nonlinear model, where the neural response is an arbitrary nonlinear function of the outputs of a set of linear filters applied to incoming stimuli. In this framework, a classic receptive field corresponds to the case where a single linear filter determines the spike probability. We ask whether it is possible to predict relative changes in the filters that would maintain optimal coding for a changing power spectrum. Optimality of encoding can be preserved by changing the linear filters so that the product of their amplitude spectra with those of the inputs remains unchanged. We tested this prediction by probing neurons in the cat primary visual cortex with white noise and natural inputs matched in luminance and contrast. To account for the fact that natural inputs are strongly non-Gaussian, which introduces differences between filters computed for the fully linear and the linear-nonlinear models, we computed filters as maximally informative dimensions. Optimal encoding arguments based on filtering in linear-nonlinear system quantitative described the changes in neural filtering at all spatial and temporal frequencies where tuning was changing. For example, at low temporal frequencies, low spatial frequencies are more common in the natural stimuli than in the white noise ensemble, and neurons become correspondingly less sensitive to those frequencies. Adaptation occurs over 40 seconds to many minutes, longer than most previously reported forms of adaptation.
ABSTRACTS Friday 29 June 2007 Computational neuroscience
27 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
01:15 p.m. – 02:15 p.m.
Speaker:
Prof Mandyam Srinivasan ARC Federation Fellow Queensland Brain Institute The University of Queensland
Bio:
Mandyam Srinivasan FRS is Professor of Visual Neuroscience at the Queensland Brain Institute of the University of Queensland. His research seeks to understand the principles of visual processing, perception and cognition in simple natural systems, and to apply these principles to machine vision and robotics. He is a Federation Fellow and in 2006 received the Prime Minister's Prize for Science. He has published over 180 papers, including several in Nature and Science.
Title:
Smart computations in small brains: Vision, navigation, perception and cognition in honeybees
Abstract:
Insects, in general, and honeybees, in particular, perform remarkably well at seeing and perceiving the world and navigating effectively in it, despite possessing a brain that weighs less than a milligram and carries fewer than 0.01% as many neurons as ours does. Although most insects lack stereo vision, they use a number of ingenious strategies for perceiving their world in three dimensions and navigating successfully in it. For example, distances to objects are gauged in terms of the apparent speeds of motion of the objects' images, rather than by using complex stereo mechanisms. Objects are distinguished from backgrounds by sensing the apparent relative motion at the boundary. Narrow gaps are negotiated by balancing the apparent speeds of the images in the two eyes. Flight speed is regulated by holding constant the average image velocity as seen by both eyes. Bees landing on a horizontal surface hold constant the image velocity of the surface as they approach it, thus automatically ensuring that flight speed is close to zero at touchdown. Foraging bees gauge distance flown by integrating optic flow: they possess a visually-driven "odometer" that is robust to variations in wind, body weight, energy expenditure, and the properties of the visual environment. Recent research on honeybee perception and cognition is beginning to reveal that these insects may not be the simple, reflexive creatures that they were once assumed to be. For example, bees can learn rather general features of flowers and landmarks, such as colour, orientation and symmetry, and apply them to distinguish between objects that they have never previously encountered. Bees exhibit "top-down" processing: that is, they are capable of using prior knowledge to detect poorly visible or camouflaged objects. They can navigate through labyrinths by learning path regularities, and by using symbolic signposts. They can learn to form complex associations and to acquire abstract concepts such as "sameness" and "difference. Bees are also capable of associative recall: that is, a familiar scent can trigger recall of an associated colour, or even of a navigational route to a food location. All of these observations suggest that there is no hard dichotomy between invertebrates and vertebrates in the context of perception, learning and "cognition"; and that brain size is not necessarily a reliable predictor of perceptual capacity. Finally, some of the above principles - especially those that relate vision and navigation - are offering novel, computationally elegant solutions to persistent problems in machine vision and robot navigation. Thus, we have been using some of the insect-based strategies described above to design, implement and test biologically-inspired algorithms for the guidance of autonomous terrestrial and aerial vehicles.
ABSTRACTS Friday 29 June 2007 Computational neuroscience
28 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
02:15 p.m. – 03:15 p.m.
Speaker:
Dr Anthony Bell Redwood Center for Theoretical Neuroscience The University of California, Berkeley, USA
Bio:
Tony Bell was awarded his M.A. in Computer Science and Philosophy at the University of St. Andrews in Scotland in 1986, and his Ph.D. in Artificial Intelligence from the Free University of Brussels in Belgium in 1993. He worked at Terry Sejnowski's Computational Neurobiology lab at the Salk Institute in San Diego in several periods from 1990 to 2001, during which time he also worked at Interval Research in Silicon Valley. Since 2002, he has been a research scientist at the Redwood Neuroscience Center, first in Silicon Valley, and now at the University of California at Berkeley. His current research interest is to connect machine learning ideas with processes known to occur in neurons and at synapses, specifically to show how molecular and spike timing-dependent processes may be implementing an `inter-level' unsupervised learning process. Earlier work was on `within-level' information maximisation at the spiking level (with Lucas Parra) and at the rate-coding level (with Terry Sejnowski). The latter algorithm yielded many `ICA' results: learning neural receptive fields, separating EEG signals and fMRI signals, and it has gone on to be used for analysing earthquakes, the internet, tumors and satellite images, amongst many other things.
Title:
Towards a theory of learning and levels for neurobiology
Abstract:
Learning, plasticity, adaptivity: these occur at the ecological, behavioural, neural and molecular levels amongst others. Yet each level is just a different description of the same processes. Defined structural relations exist between the levels (networks within networks), and these define causal relations in time. I will describe how these causal relations are essentially inter-level in nature, consisting of downward 'boundary conditions' and upward 'emergence'. Using them, information can travel from the top to the bottom of the reductionist hierarchy and vice-versa, a revolutionary idea for computational modelling. It opens the possibility of defining new kinds of learning algorithm which exploit inter-level mappings for representational purposes. This arises because mappings can be thought of as adaptive probabilistic models, as easily demonstrated by the Infomax/ICA algorithm (which I will use to illustrate this concept). The inter-level mapping of most interest to neuroscientists is the massively over complete neuron-to-synapse mapping. I will argue that Spike Timing-Dependent Synaptic Plasticity (STDP) is optimising this mapping, and that the informational readout is at the post-synaptic density, not the axon hillock. The talk is primarily conceptual and forward-looking, with a review of many empirical neuroscience findings, but technical machine learning ideas are explained, as the goal of this work is to connect biology and learning theory.
ABSTRACTS Friday 29 June 2007 Computational neuroscience
29 2007 Winter School in Mathematical and Computational Biology
25–29 June 2007
Session:
03:30 p.m. – 04:30 p.m.
Speaker:
A/Prof Michael Breakspear School of Psychiatry The Black Dog Institute University of New South Wales
Bio:
Michael Breakspear is Associate Professor of Psychiatry at the University of New South Wales and The Black Dog Institute, Sydney. He studies nonlinear neuronal dynamics in mathematical models of the brain and in experimental neuroscience data. He is interested in how disturbances of these dynamics may explain disorders such as epilepsy, schizophrenia and bipolar disorder.
Title:
Network structure of cerebral cortex shapes neuronal dynamics on multiple time scales
Abstract:
In the cerebral cortex, neuronal dynamics unfolding on structural connections give rise to complex patterns of neural activity, even in the absence of any external input. This presentation will overview graph-based analyses of large- scale brain architectures, and the complex structure-function relationships that emerge when nonlinear neuronal activity is simulated on them. Graph theory allows one to characterize both local and global properties of connected systems, including the cliqueshness, "path length", efficiency and robustness to damage. We start by reviewing the evidence for "small world" properties in mammalian cortex. Physiologically-based nonlinear dynamics are them simulated, with individual brain "nodes" inter connected according to the known anatomy of macaque neocortex. We find structure-function relations at multiple temporal scales. Functional networks recovered from long windows of neural activity (minutes) exhibit significant overlap with underlying structural networks. Hubs in these "slow" functional networks largely correspond to hubs in structural networks. However, the sequence of functional networks recovered from consecutive shorter time windows (seconds) exhibits significant fluctuations in overall functional topology. As the informational couplings between brain regions transiently shift, the centrality of nodes within the functional network is altered. Transient couplings between brain regions occur in a coordinated manner, producing two anticorrelated functional clusters, and the clusters in turn are linked by prefrontal and parietal regions that are hubs in the underlying structural network.
SPONSORS 2007 Winter School in Mathematical and Computational Biology is sponsored by the following organisations:
ARC Centre in Bioinformatics
ARC Centre in Bioinformatics
Institute for Molecular Bioscience, The University of Queensland
The MathWorks Australia Pty Ltd
Queensland Cyber Infrastructure Foundation
SGI
Top Related