AT A GLANCE
Staff scientist at the Jackson Laboratories, Bar Harbor, Maine
PhD in biostatistics from the University of Washington.
Lab funded by NIH grants to do microarray analysis, complex trait genetics; also has received NCI and American Heart Association grants.
Major interests: complex trait genetics, especially how genetics determines disease phenotypes such as hypertension.
QYou recently wrote about the statistical analysis of microarray data in a December Nature Genetics article, Sex, Flies, and Microarrays. In that article you discuss another group’s research comparing Drosophila gene expression across different groups of older and younger flies, male vs. female flies, and two strains of flies under two different conditions to see whether any of these factors influenced the differential gene expression in the different conditions. You discuss the researchers’ use of the ANOVA or analysis of variance statistical method, in which between-sample variation is compared to within-sample variation to see if differences between groups are significant, and suggest it could be useful in other experiments. Why do you see ANOVA as a useful tool for microarray data analysis?
AANOVA is the obvious thing to do. There are certain problems with it and we are flat out busy trying to tweak the edges. But whenever you have an experiment where the evident sources of variability include different arrays, different brightness, several samples of RNA, and different dyes, the obvious thing to do is to use a method [wherein you compare within- and between- sample variance.]
QIn the Drosophila experiment you discuss in your article, ANOVA does seem an obvious way to sort out the influence of various factors given that the researchers are comparing three different factors across two conditions. But most experiments are just looking at condition A vs. condition B. Does ANOVA apply to these garden variety microarray experiments?
AEven if you want to look at two states you can apply this method. But one thing that has concerned me since the first Science article came out [using microarrays] is that people seem very rigid in that they had to do a certain assay, using a reference sample with Cy3 dye and a treatment sample with Cy5. There is a fixed mindset that you could only do it this way.
QIn your article you suggest that researchers do a dye flip, where they reverse the Cy3 and Cy5 dyes in a replicate, using the Cy5 dye for the reference sample and the Cy3 for the treatment sample. Why do you suggest this method?
AThe two dyes Cy3 and Cy5 behave differently. Dye flipping corrects for the dye bias. I can think of several experiments where an investigator is looking at samples over time. We do multiple dye flips at each time point, and we make sure we get several different time points.
QWhen do you still use a reference sample and when do you omit a reference sample?
AThere are some good practical reasons to have a reference sample. The reference sample is likely to be handy when things are coming in kind of haphazardly, and you are doing exploratory research, and one of your goals to going to be to cluster expressed genes. The reference is less useful when you have a structure in mind, such as a treatment and control group. You might as well spend time measuring the treatments you are interested in studying.
QI understand that NIH funding is often predicated on having a hypothesis for your experiment, and consequently looking at known genes. This would seem to make it difficult to get funding for many microarray experiments that are query-driven.
AIt’s been that way in the past with NIH funding, but it depends on the study and what the big picture is. Some people have been able to get funds for exploratory work. My own funding goes to develop the statistical methods for microarray analysis. I piggyback on exploratory work [under this grant].
QWhat technological improvements in microarrays would help advance the field?
AWhen focusing on improving the technology, dyes are a good place to start. New dyes would be okay, but a deep understanding of the physics of the dyes would be more helpful, as sometimes the dyes interact with the specific clones. We see spots that are always green no matter what. Dye flipping is sort of a stopgap to help us patch up this technical issue with microarrays.