STATISTICAL INTEGRATION OF GENETIC INFORMATION ACROSS DATA DOMAINS: BIOMEDICAL, AGRICULTURAL, AND COMPARATIVE GENOMICS

Article from Birmingham Post-Herald
IMS Mini-Conference Report

Overview and Purpose:

As pointed out in a publication cosponsored by the IMS¹ virtually all of modern statistical analysis can be seen as involving the integration of multiple pieces of data. This is most obvious and explicit in what is now termed pooling or meta-analysis. In the fields of genomics research the need for integrating multiple sources of information via objective the quantitative analysis is especially explicit and acute. Genomics is often defined as the science of studying a very large number of genes or genetic influences simultaneously. Thus, by its very definition, genomics involves the integration of data across many domains where domains are diverse genes. But beyond this obvious issue there is a need to integrate data in several other genomic contexts. For example, genome scans in which an investigator searches for the effects of a single quantitative trait locus (i.e. a gene influencing the value of a quantitative trait) by conducting successive tests across the genome until significant results are observed are very expensive. Multiple genome scans have now been conducted in certain areas of study and for certain traits such as amount of body fat. Most studies are underpowered and offer insufficient precision of estimation for the QTL location. One potential way of increasing the precision of estimation is by increasing the sample size. Pooling data from multiple genome scans may effectively do this. However, doing so often creates great complexities because genome scans may be conducted in different strains of animals from the same species or using different experimental designs (i.e. an F2 cross versus a back cross in mice). Nevertheless, it may be possible to combine such data fruitfully, and a leader in this field is Dr. Chris Haley from the Roslin Institute in Scotland. Methods based on mixed model analysis and variance components approaches may be especially useful in this regard. The situation becomes even more complex when one wishes to consider data from genome scans across species. With traits such as body weight or degree of body fat, genome scans have been conducted in a wide diversity of species including laboratory models such as mouse and rat, agriculturally important species such as cattle, pigs, and chickens, and, of course, humans. Genetic investigators are quite used to looking across species to help place their results in perspective. An investigator who obtains a statistically significant linkage finding in a particular region of the human genome is more likely to believe that finding if a statistically significant result has been obtained in a syntenic region of the mouse genome and a mouse QTL scan. However, to date, this “strengthening of confidence” is done only at a subjective level. Nevertheless, it may be possible to make this objective. Methods based on Empirical Bayes may be ideally suited and Dr. Allison and his group are pursuing such methods. Yet a third issue involving integrating data occurs when data are integrated across completely disparate domains. The modern field of genomics involves not only QTL mapping in which putative QTL effects are tested at hundreds or thousands of points along a genome, but also gene expression studies in which the magnitude of expression or difference in the magnitude of expression between groups is evaluated for tens of thousands of genes across the genome. Similarly, proteomic information is becoming available on a scale of tens of thousands of genes, as is information about the sequence of those genes. It may be possible to integrate these different types of data and single analyses that help us better understand gene networks. Dr. Grier Page is currently working on methods to integrate QTL mapping with gene expression studies from microarray analysis.

[1]  Draper, D., Graver, D. P., Goel, P. K., Greenhouse, J. B., Hedges, L.V., Morris, C. N., Tucker, J. R., and Waternaux, C. M. (1992) Combining Information: statistical issues and opportunities for research. Washington, DC.: National Academy Press

Agenda:

START TIME	SPEAKER/MODERATOR	TOPIC	RESOURCE(S)
8:00 A.M.		Refreshments
8:30 A.M.	Philip Wood, DVM, Ph.D.	Perspective of Comparative Genomicist. Comparative Genomics of Diabetes.	Abstract
9:15 A.M.	Christopher Haley, Ph.D.	Integrating Quantitative Trait Locus Analyses Across Studies and Designs.	Abstract
10:30 A.M.	David B. Allison, Ph.D.	Discussion
10:45 A.M.		Coffee Break
11:00 A.M.	Carol J. Etzel, Ph.D.	Incorporation of Gene Information for Conducting Meta-Analyses.	Abstract
11:45 A.M.	Mark Beasley, Ph.D.	Discussion
12:00 Noon		Lunch Break
1:30 P.M.	David B. Allison, Ph.D.	Integrating QTL Mapping Data Across Species: Compartive Genomics Made Quantitative.	Abstract
2:00 P.M.	Varghese George, Ph.D.	Discussion
2:15 P.M.	Grier Page, Ph.D.	Combining QTL Mapping Data and Microarray Gene Expresssion Data in Joint Analyses.	Abstract
2:40 P.M.	Jode Edwards, Ph.D.	Discussion
3:00 P.M.		Coffee Break
3:15 P.M.	Nengjun Yi, Ph.D.	Bayesian Approaches for More Precise QTL Location Estimation in Genome Scans.	Abstract
3:45 P.M.	Hemant Tiwari, Ph.D.	Discussion
4:00 P.M.	Al Bartolucci, Ph.D.	Open Discussion: Questions and Answers of Any Speaker from the Audience

**NOTE: You will need RealOne Player to view videos. To install RealOne Player, please click on this image -