Help with interpreting results

[Chris Whelan]

Hi All,

I'm trying to use GenometriCorr to analyze the relationship of a set of genomic regions to a variety of genomic features in the human genome. I really like the ideas behind some of the statistical tests, like the relative and absolute distance distributions.

I'm wondering, however, about how to know when each test is appropriate for each type of data. For example, when testing a variety of features, most of them are coming up with a "scaled.absolute.min.distance.sum.p.value" of <0.001 (most with scaled.absolute.min.distance.sum.lower.tail = 1). The fact that most of the tests are so significant leads me to believe that perhaps the background model this test is using isn't appropriate for my data. What steps do people recommend taking to decide which tests in GenometriCorr to use? Are there any useful rules of thumb for looking at the results of all the different tests together to interpret the results?

Thanks,

Chris

[Alexander Favorov]

Hi Chris,
thank you for using GenometriCorr!

The result you see ("scaled.absolute.min.distance.sum.p.value" of

<0.001 (most with scaled.absolute.min.distance.sum.lower.tail = 1)

shows the absolute position colocation of your features on the
chromosome (they are correlated on the chromosome scale) . If it is
what you are interested in, it is OK. Still, it could show that all
your features tend to share the common regions (e.g. open chromatin
areas) while they are neutral to each other. In this case, you can
either

i. Look at the relative distance test results (or other tests like
Jaccard, they are sensitive to both local and chromosome-scale
correlation)
ii. Use mapping if you know where to map. The mapping will be improved
in the next version.

Cheers,
Sasha ( it is short name for Alexander :) ) .

> --
> You received this message because you are subscribed to the Google Groups
> "Genometricorr-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to genometricorr-u...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>