calculate % of features unique to a sample

[rmflight]

It seems that another indicator of problems would be the percentage of all of the features that are sample specific.

For example, if a sample has a large number of features that are only in that sample and no other, then we expect there could be a problem.

[rmflight]

@jesudk2, it looks like we had this issue with a sample in one of our analyses that we could test this on, correct?

[jesudk2]

yes there is one sample that has features that are sample specific. I can check against that sample and see what percentage there is.

[rmflight]

The idea is that I'll write a function in this pkg to do the calculations. I just wanted to confirm that we had an example in a data set to test this with. If you want to do an initial calculation to see if this metric might even be useful, that would be good.

[jesudk2]

I will work on that today. I think it will be extremely useful though
considering the blinded validation studies being performed. The more
information available for QA/QC and potential confounding issues of samples
will be good to have don't you think?

On Wed, Apr 13, 2016 at 3:46 PM, Robert M Flight [email protected]
wrote:

The idea is that I'll write a function in this pkg to do the calculations.
I just wanted to confirm that we had an example in a data set to test this
with. If you want to do an initial calculation to see if this metric might
even be useful, that would be good.

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#9 (comment)

[rmflight]

Yes, I think so. I'm also interested to see how this sample truly compares to the others, and how useful this metric will actually be. We have an indication of it given how many features disappear when we remove that sample from consideration at the beginning, but this calculation actually gives us a firm number / fraction, and we can see if it really is an outlier with respect to this value.