How to statistically test whether overlapping transcripts belong to separate genes or not

Question

[jon.brate]

Hi all,

I have a list of transcripts that I am interested in for several reasons. But many of them overlap other transcripts that are annotated as a different gene. But I am afraid that this could be misannotation and that my transcripts actually belong to these overlapping genes. I have expression data for many different samples and I was thinking that I could do a correlation test between the overlapping pairs from all the different samples. But I am not sure if this is the best procedure.

I wonder what would be the best test for this and how to do it.

Thanks

Answer 1

[karl.stamm]

No statistical test can tell you the names of things. Gene expression correlation is irrelevant, as well as confounded by various region specific factors.

Your best path is probably to research all of the ID#s you can find and see what they refer to. Sometimes one gene has many names and it takes literature review to see what the discoverers thought of each transcript.

Alternatively, ignore the overlapping genes and do all your work on the individual transcript level.

One possible problem is if your data is microarray based, that's really from a genomic probe (oligonucleotide binding) and one measurement might actually underlie several genes, or at least every gene using that bit of sequence. Genes with too much sequence similarity will have correlated expression measurements, unless the microarray designer was careful about designing unique probes.