The basics

What is a microarray? What can I do with them?
You can find some general information here.

What papers should I read to understand microarrays?
See the Related Links page.

Choosing and getting arrays

Do you have arrays that I can use?
We have arrays available for human and mouse experiments. See Products.

How much do arrays cost?
There are prices lists for spotted arrays.

What arrays are available, and what type are they?
The Genomics Core has both human and mouse spotted array services. See Products.

What genes are represented on the arrays?
For the spotted arrays, see Products.

Using arrays

Can you help me design experiments with microarrays?
The best advice we can give is to use the same methods you would use with any other kind of assay: design controls carefully, replicate your results, and treat all your samples as identically as possible.

How much RNA do I need to run an array?
See Array Hybridization.

How many replicate arrays should I do?
Hard to say, but do as many as you can afford. For spotted arrays, we do 3-5 replicates but some people do as many as 10. Whatever you do, don't do just one.

Can I use total RNA instead of poly A+?
Yes, see Array Hybridization.

Analyzing arrays

What software is available for analyzing my arrays?
GenePix is available in the Genomics Core. Contact Yan Shi for more information.

Can you help me with analyzing my arrays?
We have provided a database (UNC MD) for data storage and array analysis that should help you get through the initial stages of analysis. For more information on setting up an account and using UNC MD contact one of the microarray database curators . Currently we do not provide custom analysis services, so your best bet is to establish a scientific collaboration with members biostatistics or the informatics group at UNC.

What does it mean to 'normalize' a two-color array?
Because two-color arrays involve the use of two independent RNA samples that might vary slightly in concentration, and two separate labeling reactions which might have proceeded at different efficiencies, the overall brightness of the signal from one sample is often brighter than that from the other. If this is not corrected for, the data will be skewed to indicate that the sample which had more RNA or better labeling had higher expression levels (on average) in every spot. Normalization refers to the correction. There are at least three different methods which have been used for normalizing arrays:

1. Measure the total intensity of the two colors, and use the overall ratio as a correction factor. The correction factor is applied to each ratio. This method is based on the assumption that the average gene expression ratio on your array is 1. If your experimental manipulation is expected to cause more than 10-20% of all genes to change expression level in one direction, this method will not be reliable. The advantage of this method is that you can do it on any array. This is the only method currently implemented in our analysis software.

2. Use the expression ratios of spiked control RNAS. This procedure requires that your arrays include a number of control cDNAs which will not normally be present in your sample. For example, if you are doing a human array, some bacterial clones might be selected. Equal amounts of these RNAs are then 'spiked' into your samples. The correction factor is calculated as in the previous method, but only the control spots are used in the calculation. The drawback to this method is the extra effort it takes to prepare the controls.

3. Use housekeeping genes which are presumed not to change as controls. This is an extension of the procedure many people use when doing quantitative Northern blots - actin or GADPH are commonly used in that application. Obviously this method is only as good as your confidence that the genes you selected really don't change expression level, and for this reason most people don't use this method.

How do I know when a change in expression is 'significant'?
This is a difficult question to answer, and one that an increasing number of statisticians are investigating. The main problem in determining significance is the (typically) small number of replicate measurements. In addition, in spotted arrays we have found that most genes show a large variance in expression level, and this variability seems to be related to the state of the sample (i.e., it is biological in nature) or due to differences in sample handling.

It is important to keep in mind that analyzing array data for 'changed genes' is basically a game of deciding whether false positives or false negatives are more costly. It is difficult to provide a meaningful cutoff such as "2-fold changes are significant", because other factors must be considered such as how confident you are in the individual measurements which make up a ratio. Since the goal of most users is to find candidate genes which can then be followed up on, a better procedure is to rank genes with the assistance of some kind of confidence measure, add a good dose of biological knowledge, and take things from there.