Mining the HapMap for variation in the biology of health disparities in the African Diaspora

April 4-6, 2006

ABSTRACT

Inherited variation in the human genome has a critical, but largely uncharacterized role in human disease. The advent of genome-wide variation resources such as the HapMap opens a new era in population genetics, offering an unprecedented opportunity to investigate evolutionary forces that have shaped variation in natural populations. Population-based genome variation in single nucleotide polymorphisms (SNP) haplotypes can be informative in dissecting the biology of health disparities in the African Diaspora. This seminar relates knowledge gained from the Human Genome Project and research on human genome sequence variation to research strategies for achieving the U.S. Public Health Service goals of “Healthy People 2010”.

GEORGIA M. DUNSTON, Ph.D., is Professor and former Chair of the Department of Microbiology, Howard University College of Medicine in Washington, DC. She is also founding director of the National Human Genome Center (NHGC) at Howard University. Her molecular genetics research interest in human genome variation in disease susceptibility has been the vanguard of current efforts at Howard University to build national and international research collaborations focusing on the genetics of diseases common in African Americans and other African Diaspora populations. As program director of the Coordinating Center for the Africa America Diabetes Mellitus (AADM) Study, Dr. Dunston was instrumental in building an international collaboration to study the genetics of type 2 diabetes in ancestral populations of African Americans, and as program director of the Coordinating Center for the African American Hereditary Prostate Cancer (AAHPC) Study Network, she has been instrumental in building a national cooperative to map and characterize genes for prostate cancer in African Americans. The NHGC core research programs provide a foundation for bringing multicultural perspectives and resources together in the application of knowledge gained from the Human Genome Project and research on human genome variation to bear on “Healthy People 2010”, the national public health goals of disease prevention, health promotion, and the elimination of health disparities.

Linking the Genes: Strategies for inferring regulatory networks from data

February 14-16, 2006

ABSTRACT

The goal of bioinformatics, on the one hand, is to create resources for storage of the massive genomic data sets currently produced by many laboratories all around the world and, on the other hand, to perform analysis to extract information out of these datasets providing better understanding of biological systems, which eventually may lead to better treatment of human diseases and strategies for agricultural improvements. Because of the large number of different components (metabolites, proteins, mRNAs, and genes) and the complex interactions between them, mathematical modeling and multivariate statistical analysis will be a necessity for better understanding of biological systems.

In my presentation I will focus on the analysis of gene expression data, such as obtained from, for example, microarrays and gene chips. In particular I will describe three methods developed to extract gene network models from such data. A gene network is a network model in which the nodes are genes (or gene activities, mRNAs) and the directed edges correspond to direct influences between genes ("direct" in the sense that the influence is not mediated by any other gene). Such network models provide a description of genome wide genetic regulation and serve as a starting point of systems analysis and more detailed mathematical modeling.

Inferring the topology of gene networks rests mainly on the ability to distinguish direct from indirect causal influences between genes. The three methods I will describe do exactly that.

The first method I present is based on a systematic perturbation strategy. The method is experimentally very demanding as it requires each gene in the network under study to be perturbed individually and the gene expression responses for all genes to be measured. Having quantitatively determined the global expression responses and ordered these in a matrix, a network model can be obtained by a simple matrix inversion. Unfortunately, this method is sensitive to experimental noise. A computational extension based on regression with subset selection enables reliable network inference from noise experimental data, even in the case when not each gene in the network has been perturbed. Results of the evaluation of the method on simulated data are presented.

Many studies have focused on statistical correlation (Pearson or Spearman correlation) between gene expressions levels, mostly for dimension reduction techniques, but also for the purpose of network inference. Correlation graphs are formed by connecting any pair of gene-nodes by undirected edges whenever the correlation between them is statistically significant. However, it is widely known that such correlation graphs do not correspond to the actual underlying causal graph of the regulatory system, not only because correlations are undirected, but also because many correlations will be induced transitively (indirectly). Higher order correlation could assist in determining which of the correlations in the correlation graph are due to direct effects and which of those are indirectly caused. The second method I present uses first and second order partial correlations for this purpose. A first order partial correlation is the correlation between a pair of variables conditioned on a third. For example, if two genes are correlated because they are both regulated by the same factor, the correlation between them will disappear when conditioned on that factor. Similarly, second order correlation is a correlation conditioned on a pair of variables. Again, results of the evaluation of the method on simulated data are presented as well as results for a yeast gene expression data set.

The third method integrates gene expression data, genetic marker data and DNA sequence information. The idea behind this approach is that in a population of organisms there is genetic diversity as manifested by polymorphisms along their chromosomes. These polymorphisms may lead to slight modifications of transcription rates of genes or kinetic properties of their protein products which in turn may affect an observable phenotype. For several organisms genetic marker maps exist and these can be used to characterize this genetic diversity in a population. Using statistical analysis these genetic markers have been used to link complex phenotypes to locations in the genome (Quantitative Trait Loci). Once such QTLs have been found it can be concluded that there is a gene (or several genes) present in the QTL region involved in the causal process giving rise to the phenotype. In a similar fashion, gene expression levels could be treated as phenotypic traits and genomic positions affecting the expression levels of the genes can be discovered by QTL mapping (genetical genomics). Being in the possession of the DNA sequence of the organism under study one could zoom into the QTL region to find out which genes are located in there and so find genes which are potential regulators of the genes whose expression profile is affected by the QTL. The method I will present is based on four steps: 1) QTL analysis of gene expression profiles obtained from individuals in a population of organisms, 2) selecting which gene in the QTL region is the actual causal effector of the expression traits linked to it, 3) building an Encompassing Network that includes all causal influences (direct and indirect) from the regulator genes (located in QTL region) to the target genes (influenced by QTL region) and 4) removing the indirect causal influences to retain a network of only direct causal influences. For this purpose Structural Equation Modeling is used, a technique related to Bayesian networks, but able to deal with cyclic models. Model search within the Encompassing Network is performed based on greedy hill climbing using the Bayesian Information Criterion. I will present the results of the evaluation of the method on simulated data as well as some preliminary results for an integrated yeast data set.

BIO of SPEAKER

Alberto de la Fuente currently works as a post-doctoral researcher with the Virginia Bioinformatics Institute (VBI). Under the guidance of Dr. Ina Hoeschele, head of the Statistical Genetics group at VBI, he is developing a strategy for gene network inference by integrating gene expression data, genetic marker data and DNA sequence information. Alberto received his BSc degree in 1996 in Biotechnology at the College of Amsterdam. He continued his education at the University of Amsterdam, where he developed interests in quantitative approaches to molecular biology and biochemistry. He received training in mathematical modeling of biological systems, metabolic control analysis and biophysics. In 1998 he completed his MSc degree in General Biology. In pursuit of his interest in theoretical and computational approaches to biochemistry, he started his doctoral research in 2000. Alberto will obtain his doctoral degree from the Free University of Amsterdam, but he carried out his research at VBI. His advisor, Dr. Hans Westerhoff, is a leading figure in Systems Biology in Europe. At VBI, he worked under the supervision of Dr. Pedro Mendes, author of the biochemical simulation software called Gepasi. During his initial time at VBI, he focussed on biochemical network simulation and analysis of gene expression and metabolomics data. Among several contributions to the field, he developed a quantitative analysis to infer gene networks from gene expression data using an experimental perturbation strategy and a method using the concept of partial correlations for network inference. Alberto will defend his doctoral dissertation in Amsterdam in July 2006.

PathoSystems Biology, Cyberinfrastructure and Overview
of the Virginia Bioinformatics Institute (VBI)

November 15-17, 2005

ABSTRACT

Infectious diseases affect animals and plants, including humans, that are important to our society at diverse levels. Life Scientists interested in studying infectious diseases have historically been segregated into disciplinary fields, such as microbiology, epidemiology, immunology, etc., and there has been little interaction and communication between those fields. In addition, very few life scientists working on infectious have had access to or made use of diverse new and powerful computational and laboratory technologies to allow new ways to develop vaccines, therapeutics, diagnostics and other coutermeasures against infectious agents. We have coined the term PathoSystems Biology to express the use of diverse high performance laboratory techniques (genomics, transcriptomics, proteomics and metabolomics, for example, coupled with modeling, simulation and theory development and application to infectious diseases, in an integrated, team-oriented fashion that characterizes systems-level approaches. We have also built and cyberinfrastructure platform within VBI that brings together diverse components, infrastructure, and people to the study of infectious diseases.

BIO of SPEAKER

Dr. Sobral is currently the Executive and Scientific Director of the Virginia Bioinformatics Institute (VBI) at Virginia Tech in Blacksburg, Virginia; and works also as Adjunct Research Professor of cancer Biology in the Comprehensive Cancer Center at Wake Forest University in Winston-Salem, North Carolina.

His scientific training began at the Department of Plant Anatomy of the Universidade Federal de Viçosa at Minas Gerais, Brazil, where he completed a B.Sc. in Agronomic Engineering. Dr. Sobral holds a Ph.D. in Genetics from Iowa State University and a Postdoctoral Degree from the California Institute of Biological Research where he trained in Molecular Evolution, Physical and Genetic Mapping.

Throughout his scientific career, his research has focused on combining computational and laboratory technologies for the improvement of diagnostics and the development of new vaccines and therapeutics. With his team of collaborators at VBI, Dr. Sobral has developed a new approach for the study of infectious disseases which they have named PathoSystems Biology. This novel approach combines the fields of genomics, transcriptomics, proteomics and metabolomics with modeling, simulation and theory development applications. At VBI, they have succeeded in building a cyber-infrastructure platform that makes possible to bring all these components together.

SCHEDULE of ACTIVITY