FC calculation in Limma
1
0
Entering edit mode
Lana Schaffer ★ 1.3k
@lana-schaffer-1056
Last seen 10.3 years ago
Unfortunately, we have become aware recently that the FC calculation in Limma Is not correct. We input the expression values as log2 transformed values. The Average log expression value is then calculated by averaging the log values (ave(logX)). However, this is NOT the log of the average expression value (logXbar) and so the reported FC is therefore not correct. Is there a possibility that the code can be corrected? Lana Schaffer Biostatistics, Informatics DNA Array Core Facility 858-784-2263 [[alternative HTML version deleted]]
• 1.5k views
ADD COMMENT •
0
Entering edit mode
@kasper-daniel-hansen-2979
Last seen 17 months ago
United States
What is wrong with this? Many papers have shown that (some) noise is additive on the log scale, so the relevant calculation should be ave(log(X)) Indeed, using log(ave(X)) goes against essentially the entire field of microarray analysis. Of course, there might be special cases, but it seems you are making a general compliant. So you need to come with some convincing references to convince me (us) that there is merit in your statement. (of course, I am not a limma author, so my opinion caries little weight). Kasper Postdoc, Department of Biostatistics, Johns Hopkins University On Tue, Jan 18, 2011 at 7:19 PM, Lana Schaffer <schaffer at="" scripps.edu=""> wrote: > Unfortunately, we have become aware recently that the FC calculation in Limma > Is not correct. ?We input the expression values as log2 transformed values. ?The > Average log expression value is then calculated by averaging the log values (ave(logX)). > However, this is NOT the log of the average expression value (logXbar) and so the reported > FC is therefore not correct. > Is there a possibility that the code can be corrected? > > Lana Schaffer > Biostatistics, Informatics > DNA Array Core Facility > 858-784-2263 > > > ? ? ? ?[[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >
ADD COMMENT •
0
Entering edit mode
Kasper, How do you tell customers that the fold-change is not the Average of X divided by the average Y signal? Lana -----Original Message----- From: Kasper Daniel Hansen [mailto:kasperdanielhansen@gmail.com] Sent: Tuesday, January 18, 2011 5:34 PM To: Lana Schaffer Cc: bioconductor at r-project.org Subject: Re: [BioC] FC calculation in Limma What is wrong with this? Many papers have shown that (some) noise is additive on the log scale, so the relevant calculation should be ave(log(X)) Indeed, using log(ave(X)) goes against essentially the entire field of microarray analysis. Of course, there might be special cases, but it seems you are making a general compliant. So you need to come with some convincing references to convince me (us) that there is merit in your statement. (of course, I am not a limma author, so my opinion caries little weight). Kasper Postdoc, Department of Biostatistics, Johns Hopkins University On Tue, Jan 18, 2011 at 7:19 PM, Lana Schaffer <schaffer at="" scripps.edu=""> wrote: > Unfortunately, we have become aware recently that the FC calculation in Limma > Is not correct. ?We input the expression values as log2 transformed values. ?The > Average log expression value is then calculated by averaging the log values (ave(logX)). > However, this is NOT the log of the average expression value (logXbar) and so the reported > FC is therefore not correct. > Is there a possibility that the code can be corrected? > > Lana Schaffer > Biostatistics, Informatics > DNA Array Core Facility > 858-784-2263 > > > ? ? ? ?[[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >
ADD REPLY •
0
Entering edit mode
If they are savvy I would explain that the log scale is the relevant scale for analysis of microarrays and that they should therefore start by log transforming. A corollary to this is (of course) that fold changes are calculated as they are. I would explain that this has to do with the distribution of noise in the intensity measurements. If they are less savvy I would make an authoritative argument and would say that this is how it has been done in the field in the last 10 years or more. If this does not help, I would count myself lucky that they are not my collaborators. Of course, that does not help you. Kasper On Tue, Jan 18, 2011 at 9:20 PM, Lana Schaffer <schaffer at="" scripps.edu=""> wrote: > Kasper, > How do you tell customers that the fold-change is not the > Average of X divided by the average Y signal? > Lana > > -----Original Message----- > From: Kasper Daniel Hansen [mailto:kasperdanielhansen at gmail.com] > Sent: Tuesday, January 18, 2011 5:34 PM > To: Lana Schaffer > Cc: bioconductor at r-project.org > Subject: Re: [BioC] FC calculation in Limma > > What is wrong with this? ?Many papers have shown that (some) noise is > additive on the log scale, so the relevant calculation should be > ?ave(log(X)) > Indeed, using log(ave(X)) goes against essentially the entire field of > microarray analysis. ?Of course, there might be special cases, but it > seems you are making a general compliant. ?So you need to come with > some convincing references to convince me (us) that there is merit in > your statement. > > (of course, I am not a limma author, so my opinion caries little weight). > > Kasper > > Postdoc, Department of Biostatistics, Johns Hopkins University > > On Tue, Jan 18, 2011 at 7:19 PM, Lana Schaffer <schaffer at="" scripps.edu=""> wrote: >> Unfortunately, we have become aware recently that the FC calculation in Limma >> Is not correct. ?We input the expression values as log2 transformed values. ?The >> Average log expression value is then calculated by averaging the log values (ave(logX)). >> However, this is NOT the log of the average expression value (logXbar) and so the reported >> FC is therefore not correct. >> Is there a possibility that the code can be corrected? >> >> Lana Schaffer >> Biostatistics, Informatics >> DNA Array Core Facility >> 858-784-2263 >> >> >> ? ? ? ?[[alternative HTML version deleted]] >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor at r-project.org >> https://stat.ethz.ch/mailman/listinfo/bioconductor >> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >> >
ADD REPLY •
0
Entering edit mode
Hi Lana > On Tue, Jan 18, 2011 at 9:20 PM, Lana Schaffer<schaffer at="" scripps.edu=""> wrote: >> Kasper, >> How do you tell customers that the fold-change is not the >> Average of X divided by the average Y signal? On 01/19/2011 04:32 AM, Kasper Daniel Hansen wrote: > If they are savvy I would explain that the log scale is the relevant > scale for analysis of microarrays and that they should therefore start > by log transforming. A corollary to this is (of course) that fold > changes are calculated as they are. I would explain that this has to > do with the distribution of noise in the intensity measurements. > > If they are less savvy I would make an authoritative argument and > would say that this is how it has been done in the field in the last > 10 years or more. > > If this does not help, I would count myself lucky that they are not my > collaborators. Of course, that does not help you. And if you want to sidestep the argument by giving your customers another way of looking at it, you might say: "Of course, the fold change is the average of X divided by the average of Y, but it is important to use the geometric, not the arithmetic average." If you put it this way, the question is what averaging method to use, not what scale to average on. Of course, this is mathematically equivalent (the log of the geometric mean is equal to the arithmetic mean of the logs), but maybe more palpable to people who are scared of logarithms. Have fun with your customers ;-) Simon
ADD REPLY •

Login before adding your answer.

Similar Posts
Loading Similar Posts
Traffic: 703 users visited in the last hour
Content Search
Users
Tags
Badges
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6