Expression arrays are proving to be useful to address a variety of biological questions. "Optimal" experimental design varies depending on the objectives of the study, types of samples, costs of samples and arrays, and contributions of biological and technical sources of variability. In studies with clinical objectives, microarrays have been used exploratorily, to screen for candidate genes or candidate gene clusters that can be studied further by comparison to outcome and validated by alternative methods such as q-RT-PCR or IHC in independent clinical samples. More supervised methods of gene selection may better identify clinically relevant genes, but may also pose serious problems with overfitting. Study design is also difficult because there is little data on which to base sample size calculations. However, new predictive markers that are merely statistically significant, and cannot separate individual patients into therapeutically distinct groups, will not be useful. One strategy may be to specify large target effect sizes, and plan relatively small screening studies to detect genes that provide such discrimination, followed by comfirmatory studies of the selected genes.
Copyright. All Rights Reserved.