Interpretation of Picard's MEDIAN_CV_COVERAGE

[Lien]

Dear all,

We just performed an RNA-seq experiment from human samples. To get some idea of the quality, I ran Picard's CollectRNASeqMetrics.
In the output, I find 'MEAN_CV_COVERAGE'. The explanation for this value is also on the website: the median CV of coverage of the 1000 most highly expressed transcripts.

I'm not really sure what CV means. And how can I interpret this value? The value I get is 0.48. If I think of this as 48x coverage, this seems really a lot to me. Especially since visualisation with IGV shows me a lower coverage.

Any help will be appreciated!

Thanks
Lien

[Bukowski]

Dear all,

We just performed an RNA-seq experiment from human samples. To get some idea of the quality, I ran Picard's CollectRNASeqMetrics.
In the output, I find 'MEAN_CV_COVERAGE'. The explanation for this value is also on the website: the median CV of coverage of the 1000 most highly expressed transcripts.

I'm not really sure what CV means. And how can I interpret this value? The value I get is 0.48. If I think of this as 48x coverage, this seems really a lot to me. Especially since visualisation with IGV shows me a lower coverage.

Any help will be appreciated!

Thanks
Lien

I always assumed CV referred to the coefficient of variation:

http://en.wikipedia.org/wiki/Coefficient_of_variation

As in 'mean coefficient of variation of coverage'

Although happy to be proved wrong.

[jstjohn]

from the javadoc

According to the javadoc:
"The median CV of coverage of the 1000 most highly expressed transcripts. Ideal value = 0."

What is confusing me though is it looks like this involves the coefficient of variation (sample sd)/(sample mean). Since this number is reported for a single RNA-seq bam file, I am wondering where the population is coming from? It looks like the only way to get at an sd or mean from the above description is by looking at the coverage of the top 1000 transcripts. That would give you a single number, not something you could take the median of. Pretty sure I am missing something here.


http://picard.sourceforge.net/javadoc/net/sf/picard/analysis/RnaSeqMetrics.html#MEDIAN_CV_COVERAGE

[pengchy]

What use of MEDIAN_CV_COVERAGE

According to the javadoc:
"The median CV of coverage of the 1000 most highly expressed transcripts. Ideal value = 0."

What is confusing me though is it looks like this involves the coefficient of variation (sample sd)/(sample mean). Since this number is reported for a single RNA-seq bam file, I am wondering where the population is coming from? It looks like the only way to get at an sd or mean from the above description is by looking at the coverage of the top 1000 transcripts. That would give you a single number, not something you could take the median of. Pretty sure I am missing something here.


http://picard.sourceforge.net/javadoc/net/sf/picard/analysis/RnaSeqMetrics.html#MEDIAN_CV_COVERAGE

According to the definition, MEDIAN_CV_COVERAGE is "The median CV of coverage of the 1000 most highly expressed transcripts. Ideal value = 0." I think the variation should be very large, because there will be several gene expressed at very high value. Then, why the ideal value is "0"? "0" means the variation is zero, it not consistent with the fact.