Microarray Data Analysis

Designing Your cDNA Microarray Experiment

Proper statistical design is essential to get the most out of microarray experiments and ensure that the effects of interest to biologists are accurately and precisely measured. Here is a list of suggested information which a bioinformatician is likely to want to know before designing your experiment. For more information see Yang and Speed, Design issues for cDNA microarray experiments, Nature Reviews Genetics 3, August 2002, 579-588.

Information needed

1. Aim of the experiment What are the broad aims of the project? Examples, searching for differentially expressed genes, searching for patterns or discriminating between different samples. Issue to consider together are:
   1. Results from previous studies (e.g. Pilot study). This will provide an idea whether we expect a few or many changes between genes expression.
   2. Verification method. Specify the types of verification methods you will use including RT-PCR, Northern/Western analysis, In-situ hybridization, knock-out studies etc.
2. Biological background. How will these microarray experiments contribute to the overall research plan? Provide simple information and references to the biological background for this project.
3. Types of samples. What mRNA are you planning to use? (Tissue samples, cell lines etc.)
4. Specific questions / most important comparison. In reference to the types of samples listed in 2, what are the specific questions you wish to address? In addition, specify the priority the different questions.
5. The number of slides available.
6. Amount of material. Denote the amount of mRNA involved in one channel of hybridization as a unit. Specify the maximum number of units available for hybridization within a certain time period. (Example: cell lines produce limitless material, but mRNA for 1 sample takes 2 weeks)
7. The experiment process prior to hybridization. (i.e. Isolation from tissue culture -> labelling -> hybridisation). We believe the experimental process between different replicates should be as independent as possible.
8. What controls are there? Specify the controls and differentially expressed genes that are known in the comparisons you are making on the slides.

Common questions for the statistician

What types of replication can be used? The type of replication to be used in a given experiment depends on the precision of conclusions desired, and on the generalisability of the experimental results sought by the experimenter.

The number of replicates. This depends on four components; the variance of individual measurements, the magnitude of the effect to be detected, the acceptable false positive rate, the chance of detecting an effect of at least the desired magnitude (power). Since most of this information is unknown, it is difficult for the experimenter to access the desired number of replication needed, therefore in practice we advise to replicate as much as possible. (Example: for detecting genes with 1.5 fold change requires more replication than for detecting genes with >2 fold change)

The use of dye-swaps. We recommend changing the orientation of dye labelling as it reduces systematic colour bias.

Possible data analysis: t-statistics, linear models analysis, discrimination, clustering etc.

A Case Study

We ask the above questions of the experiment considered in Lin, D. M., Yang, Y. H., Scolnick, J. A., Brunet, L. J., Peng, V., Speed, T. P., and Ngai, J. (submitted). A spatial map of gene expression in the olfactory bulb. Department of Molecular and Cell Biology, University of California, Berkeley.

1. Aim: Identify spatially differentially expressed genes in the olfactory bulb.
   1. Prior results:
   2. Verification method: In-situ hybridization on P0 mouse olfactory bulb.
2. Biological background: How olfactory neurons are able to identify their glomerular targets in the bulb is unknown. One possibility is that differentially expressed molecules in the bulb help olfactory neurons distinguish regions of the bulb from one another. Possible candidates involved in this process have been extremely difficult to identify. We plan to take look at olfactory bulbs dissected from newborn, postnatal day 0 CD-1 mice (P0). Each bulb was cut into three slices along a single axis- AP, ML, or DV. Slices corresponding to the extreme ends of each axis were separately rinsed in PBS and immersed in liquid nitrogen.
   Background references, Ressler et. al. Cell 79, 1245-1255 (1994) and Lin et. al. Current Opinion Neurobiology 9, 74-78 (1999).
3. Samples: Six samples are A = anterior, P= posterior, M=medial, L=lateral, D=dorsal and V= ventral.
4. Specific Question:
   1. Identify genes that are differentially expressed or show spatial restriction in the olfactory bulb.
   2. Identify possible differentially expressed patterns in the olfactory bulb.
   3. Most important comparisons: A vs P, M vs L and D vs V.
5. Number of slides available: approximately 20 to 30 slides.
6. Amount of material: approximately 5 samples per region. No time constraint, samples collected over 2 months.
7. Experimental process: Dissection / RNA isolation -> amplification -> labelling -> hybridization. Each isolation will be made from a pool of ~10 mice and then amplified. This will produce approximately 3 to 4 samples of RNA ready for labelling. All RNA samples were labelled independently and hybridized.
8. What controls are there: None.