|
In the space of just a few years, DNA microarray technology has become one of the foremost tools in biological research. The emergence of this technology has empowered researchers in functional genomics to monitor gene expression profiles, thousands of genes (perhaps even an entire genome) at a time. Of course, this means lots of data! And, in fact, the large volume of data generated by these experiments, and the challenge their analysis presents, has created an opportunity for some very interesting statistical work.
A typical microarray experiment is performed as follows. Suppose that a scientist wishes to know which genes were differentially expressed in several different types of cells. First, s/he would extract mRNA from several samples of such cells, with the knowledge that only mRNA corresponding to any genes that were expressed would be present in these samples. S/he will reverse transcribe the mRNA to cDNA (for stability) and label the samples by adding some fluorescent dye to them. A DNA microarray for this experiment would be a tiny glass slide on which several genes of known sequence had been printed in a rectangular array of spots, one gene per spot. S/he will now expose a set of microarrays to the labeled samples, one microarray per sample, and allow enough time for any cDNA in the samples to recognize and hybridize to its complementary sequence on the microarray. S/he would then wash off any extraneous probe from the microarrays. Each spot on the microarrays where the sample bound to the spot would identify a gene that corresponds to some reverse transcribed mRNA in the sample. Such spots can be easily recognized as they are the only ones that will fluoresce. S/he will finally scan the microarrays and measure the intensity level of fluorescence at each spot. By comparing these intensities across the microarrays, s/he will know which genes are differentially expressed in the various different types of cells.
This may sound easy, but it has become evident to researchers in the field that obtaining precise and reproducible results is not. This is due to many reasons: e.g., very little biological replication, high within array variability, nonlinear between array variability, high experimental variability, sporadic outliers, spatial outliers, nonnormality, within gene heteroscedasticity, between gene heteroscedasticity, correlation among genes, multiplicity, etc.
A typical analysis involves a preprocessing stage and a processing stage. In the preprocessing stage, microarrays are checked for quality, systematic effects are removed by nonlinear normalization and data are screened for outliers and other unusual features. At the processing stage, the preprocessed data are analyzed to pick up gene expression patterns that differ significantly across the different cell types. This can be done using a gene-by-gene approach with conventional statistical methods or with more sophisticated tools that incorporate strategies for pooling information across genes, either through modeling, Bayesian-type arguments or supervised clustering.
In this short course, we intend to cover many of the statistical considerations involved in a typical microarray data analysis. We will do this mainly in the context of a single microarray dataset, through which we will illustrate the statistical issues involved at both stages of the analysis and many of the more promising statistical approaches.
Fees: M-$200 ($270 after July 18), NM-$260 ($330 after July 18), SM- $125 ($200 after July 18)
|
= Applied Session,
= Theme Session,
= Presenter