Thanks in advance, Yves -------- Original Message -------- Subject: Bioinformatics for Agribusiness Date: Wed, 22 Mar 2006 17:38:59 +1100 From: <David.Mitchell@csiro.au> To: <yvdp@psb.ugent.be> Dear Yves, Koen Bruynseels from CropDesign suggested that I drop you a line. I am the leader of the Biotechnology and Health Informatics program in CSIRO's Division of Mathematical and Information Sciences. My background is in molecular biology, but I now have the pleasure of leading a group of about 40 statisticians, mathematicians and image analysts. The group works in three main areas, bioinformatics for human health and agribusiness, imaging in biotechnology and health informatics. I've written more about our research in detail below. My colleague David Lovell (who heads up our Bioinformatics for Agribusiness group) and I will be in Gent on Tuesday 4th of April visiting CropDesign. Koen suggested that a visit to the Plant Systems Biology Department at VIB and gave me your email. As a group we are interested in Systems Biology but are still discussing how to really get some engagement. If you have some time on Tuesday afternoon, would it be possible to come for a chat? I would be happy to give a short presentation on some of our statistical bioinformatics stuff, particularly the network building. Kind Regards David ------------------------------------------------------------------------------------ CSIRO is Australia's national science agency and one of the largest in the world. CSIRO research delivers solutions for agribusiness, energy and transport, environment and natural resources, health, information technology, telecommunications, manufacturing and mineral resources. The Division of Mathematical and Information Sciences consists of mathematicians and statisticians with a focus on smarter information use in Biotechnology and Health, the Environment, and Decision Technologies. CSIRO Bioinformatics are specialists in the analysis of the types of data being generated by the current range of high throughput platforms. Our initial focus was on expression microarray data, where the number of measurements (genes) per sample vastly exceeds the number of samples. We call this type of data p>n data, because in statistics p often refers to the number of measurements (or variables) and n refers to the number of samples. Of course, p>n data occurs in many other settings and we are now active not only in expression microarrays (both Affymetrix and cDNA types) but in DNA (SNPs), proteomics (including LC/MS) and metabolomics. Our proprietary technology is focused mainly in the area of classification of samples using p>n data. For example, seeking simple combinations of gene expression parameters that are diagnostic of disease or finding combinations of SNPs that are indicative of a certain predisposition. The philosophy of our technology is multivariate; in particular, we generate classifiers using combinations of small numbers of genes, but all genes are candidates not just those preselected by univariate differential expression measures. There are two main threads to our technology; one based on the use of sophisticated Bayesian priors and an EM algorithm[1] <BLOCKED::outbind://20/#_edn2>, and the other a highly efficient variant of more a traditional stepwise approach. The Bayesian approach is able to generate sparse predictive models of complex phenotypes, including classification, survival and quantitative traits. We have also used our Bayesian approach to build "sparse" networks of genes. We have found this to be particularly useful as we are able to cross validate the linkages to build a robust network (Poster attached). While much of our work has been on data from medical applications, we have a group working on agribusiness applications as well. They have been looking at 2 phase designs for microarray experiments, building a LIMS database for managing rice breeding and looking at various problems using mathematical modelling approaches. We are also just starting work in objective measurement of plant phenotypes using imaging. ------------------------------------------------------------------------ [1] <outbind://20/#_ednref2> (“A Bayesian approach to variable selection when the number of variables is very large”, in /Science and Statistics: A Festschrift for Terry Speed/, volume 30 of Lecture Notes - Monograph Series, Institute of Mathematical Statistics, Hayward, California, 2003, Kiiveri, H. T.). David Mitchell PhD, MEI Research & Business Leader Biotechnology & Health Informatics CSIRO Mathematical & Information Sciences Locked Bag 17 North Ryde, NSW, 1670 AUSTRALIA T: +61 2 9325 3256 F: +61 2 9325 3200 M: +61 402 304 801 E: david.mitchell@csiro.au <mailto:david.mitchell@csiro.au> -- Yves Van de Peer, PhD. Professor in Bioinformatics and Genome Biology Department of Plant Systems Biology Ghent University Technologiepark 927 B-9052 Ghent Belgium Phone: +32 (0)9 331 3807 Cell Phone: +32 (0)476 560 091 Fax: +32 (0)9 331 3809 email: yves.vandepeer@psb.ugent.be http://www.psb.ugent.be/bioinformatics/