[BioC] fitPLM query

[Ben Bolstad]

You might want to consider using NUSE and RLE for quality assessment
purposes. NUSE in particular is highly dependent on the weights, with
lower weights leading to extreme NUSE values, and is a bit more
sophisticated then what you are doing now. 

Note that the absolute magnitude of the weights is of little importance.
In fact it is possible to get all weights under one if you use one of
the non-default weighting schemes implemented in fitPLM (though I doubt
anyone is actually using those). Instead it is the relative differences
in these weights, and whether the probes for these lower weights make up
a significant proportion of the probes in fall in a probeset, rather
than just stray probes here or there, that is important.

Best,

Ben



On Mon, 2007-07-23 at 16:12 -0400, dhruv pant wrote:
> Hi,
> I had a question on using fitPLM to determine quality of chips.
> 
> I run fitPLM and obtain the weights etc. from the pset object.
> >From this, I compute the percentage of probes with weight less than one for
> every chip,  and if that percentage is greater than say 20%  (empirically
> determined), I consider the chip as failing quality.
> So out of 10 chips, I get say 8 good chips.
> 
> Now, if I take these 8 good chips from the first run and do a fitPLM again
> and I use the percentage of weights idea above to determine pass/fail, I get
> a few of these good chips to fail again. It seems that for a given set of
> chips, there are always probes which will have weights less than one.
> So I was wondering if this is a good measure for quality testing.
> 
> I looked at the distribution of the residuals over all chips from the first
> run and the distribution from the second run and the spread in the second
> distribution is much less, as expected.
> 
> Thanks for any help/suggestions.
> 
> Dhruv
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor