10:00 AM Applications of cDNA Microarrays to Studies in Wood Formation: Y.-H. Sun, North Carolina State University
11:30 AM Data Analysis and Modelling of DNA Microarrays in Genetic Expression Profiling: M. West, R. Spang and H. Zuzan,
12:30 PM Lunch
1:15 PM Continuation of 11:30 Topic - Data from Affymetrix GeneChip Images: Harry Zuzan
3:15 PM Next Steps
4:00 PM Adjourn
Applications of cDNA Microarrays to Studies in Wood Formation
North Carolina State University
New technologies allowing highly parallel, high throughput genetic and molecular biology analysis have created the field of "Genomics" and "Bioinformatics". DNA Microarrays are one of these technologies that allow scientists to obtain gene expression profiles in a highly parallel way. The result is a huge volume of data generated much faster than scientists can analyze it. Not all data are of optimal quality, and some filtering is necessary before beginning of experimental analyses. Statistical methods for removing systematic errors in raw data allow useful experimental analyses to be conducted on data that would otherwise be unsuitable for analysis.
A cDNA microarray approach was used to study the relationship between changes in gene expression due to environmental stresses, and those due to developmental transitions such as maturation. We profiled expression patterns of 3107 loblolly pine genes in developing xylem of three individuals at juvenile and mature stages, and compare that to expression patterns in two year old seedlings which had been subjected to heat, water and bending stress treatments. A local regression approach was used for transforming and normalizing microarray data.
Cluster analyses of the transformed data put functionally related genes into the same clusters. Correlation studies showed that gene expression patterns of heat and water stresses are well correlated. Juvenile wood formation on the other hand showed reverse relationships with water and heat stresses, with the exception of the small heat shock proteins which were highly induced by heat stress while only moderately induced by water stresses. These results imply that juvenile wood formation is under a distinct genetic control. No apparent relationship between mechanical bending stress and juvenile wood formation was observed.
The application of local regression method is successful in recovering useful information from sub-optimal data. These biological conclusions could not have been reached based on the raw data due to systematic errors related to technical difficulties.
Data Analysis and Modelling of DNA Microarray Data in Genetic Expression Profiling
Mike West, Rainer Spang and Harry Zuzan
We report on our studies of a range of issues of data quality, exploration, summary and formal statistical analysis in genetic expression profiling using DNA microarray (oligonucleotide) arrays. Our work forms part of a collaborative project concerned with the development of relevant and appropriate statistical approaches for clinical and physiological phenotyping using array data.
These talks will discuss questions of definition of summary measures of genetic expression from DNA microarrays, issues of imaging and data access, and concerns about summary expression measures generated using standard commercial software. Statistical issues arising in studies of patterns of variation in expression of large numbers of genes in one or two applications (e.g., in cell cycle studies) highlight these issues. In connection with more formal inference in clinical and physiological studies, challenging questions of modelling and analysis arise due to the high-dimensionality of the gene expression profile. In the context of a pilot breast cancer phenotyping project we discuss one of our lines of research in this area. This is based on Bayesian analyses of binary regression models with very many more covariates/predictors than observatons, in which we use standard singular-value regression ideas coupled with novel prior models to enable formal analysis. We describe this framework and illustrate its application in both predictive discrimination of clinical outcomes and in identifying candidate genes and subsets of genes for further study.
This work is collaborative with the National Institute of Statistical Sciences, together with Drs. Joseph Nevins, Jeff Marks and Seiichi Ishida of the Duke School of Medicine, and is part of a project run under the auspices of the Duke Center for Bioinformatics and Genome Technology.