Work at SourceForge, help us to make it a better place! We have an immediate need for a Support Technician in our San Francisco or Denver office.
> I am an new user to arrays and MEV and am trying to analyse an Affy ST 1.0 array.
> By way of background, here is how we have set up the experiment.
> We are looking at whether expression of a protein (X) alters gene expression. The gene encoding this protein is stably transformed into our cell line and the X protein is expressed when tetracyline is removed from the cell media. We have another cell line transformed
with a gene expressing a different protein (Y), that is also repressed by tetracylcine but should not have the same effect on cellular gene expression. As a control, we have the non-transformed parent cell line.
> We have "tet off" and "tet on" for each cell line.
> Ie X tet off, X tet on
> Y tet off, Y tet on
> Parent tet off, parent tet on.
So I am simply trying to determine if the X-tet off cells differentially express genes relative to the other samples. Ie does expression of the X protein alter mRNA expression differently to the Y protein, and negative parent control?
> So in effect, I wish to compare one sample (X, tet off) to all the others and work out which genes are significiantly differentially expressed at say a two fold difference. Yet SAM and the other statistical programs appear to require at least 2 samples in each group?
> This is where I am stuck at present. Is it possible to compare X "tet off" to one or more of the other samples, ie one sample compared to multiple samples?
> Also, I can't actually see the gene names, just a whole lot of numbers.
> Does this mean I have not imported the annotation correctly? The file I imported, after downloading from the Affy website, is a .csv file.
>Thanks folks, Peter.
I hate to break it to you, but you really should have more than one sample per experimental condition. Microarray experiments need replicates just as much as other experiments do. Otherwise you really have no idea whether the difference in expression level between your condition of interest and your control is due to normal technical, technical error, or real biological effect.
From a statistical standpoint, as well, replicates are a requirement. The TTEST and SAM modules won't work with only one datapoint. That's not a design choice on our part, either (we're not trying to *make* you run replicate arrays), but rather a problem with the statistics of the test. In order for you to do a t-test in *any* software tool you need to calculate a variance across your samples in a give experimental condition, and since you have only one sample your variance is zero. Dividing by that variance then becomes a problem.
You can try using MeV's data filters (Adjust Data -> Data Filters -> Variance Filter) to find genes that vary between your experimental conditions.
You mentioned in email that you would eventually have some timecourse data. That's very interesting, because it means that a new module we've developed, called BETR, may be just what you're looking for. Once the rest of your data comes in you should definitely try it out. It's designed to pull differentially expressed genes out of timecourse data in a more sensitive way than standard tools, like TTEST or SAM can. There's a full description of the algorithm in the MEV manual, and our group is working on getting a paper accepted on it as well.
If MeV has loaded your annotation correctly, you should be able to choose which annotation type to display by selecting the menu Display -> Gene/row labels.
Hope this helps,