Hi Rohan, What I suspect is some "outliers" or extreme values that screw up the average fold-change, especially when you have uneven design (different # of samples in two group), for example for given gene , its readout under class A (4 samples) are: 100,300,60,80, and under class B ( 2 samples) are: 120, 140 => average fold-change (sample1/sample2)=1.038 However, it is likely that 300 value is an outlier since it is not consistent with other three meaurement, therefore this gene is likely downregulated in class A compared with class B though the average FC >1. In fact this is the power of rankproduct method as it is much more robust against outlier than t-like statistics which will call this gene up-regulated in class A. If you prefer, you would send me the expression of such genes over these 20 samples so that I would take a look ? Best, Fangxin Rohan M wrote: > Dear Fangxin, > > Thank you very much for looking into mail and replying promptly, I > really appreciate it. > For answers to question 2 and 3, I understood the points. I cross > checked that the data was normalized using MAS algorithm. > > Regarding question 1 here are few more details - > I'm using two class data with Single origin to calculate Rank product. > For example, Class A contains 20 samples and Class B contains 10 > samples. I'm using following cutoff for the topGene functionality. > > /classes - The vector to assign two classes labels./ > /RP.out <- RP(arab, classes, num.perm = 100, logged = TRUE,na.rm = > FALSE, plot = FALSE, rand = 123) > outPut <- topGene(RP.out, cutoff = 0.05, method = "pfp", logged = > TRUE, logbase = 2, gene.names = rownames(arab))/ > > Is the cutoff 0.05 enough or need to be more stringent? Also, with > this set up is it possible to have FC value more than 1 in Table1 or > FC value less than 1 in Table2 for many probes? > > Thank you once again. > > Regards, > Rohan > > On Tue, Dec 8, 2009 at 12:16 AM, Fangxin Hong > <fxhong at="" jimmy.harvard.edu="" <mailto:fxhong="" at="" jimmy.harvard.edu="">> wrote: > > Hi Rohan, > Please see my comments below, > > > > > -----Original Message----- > From: bioconductor-bounces at stat.math.ethz.ch > <mailto:bioconductor-bounces at="" stat.math.ethz.ch=""> > [mailto:bioconductor- <mailto:bioconductor-> > bounces at stat.math.ethz.ch > <mailto:bounces at="" stat.math.ethz.ch="">] On Behalf Of Rohan M > Sent: Monday, December 07, 2009 6:55 AM > To: bioconductor at stat.math.ethz.ch > <mailto:bioconductor at="" stat.math.ethz.ch=""> > Subject: [BioC] Query related to topGene functionality in > RankProd > library. > > Dear Sir, > > I'm using RankProd library to find the significant genes > in microarray > studies. I'm facing some problems in understanding the > output of > topGene > functionality. > Could you help me on following queries? > > 1) The topGene functionality outputs > Table1: Genes called significant under class1 < class2 (Up > regulated > Genes) > and Table2: Genes called significant under class1 > class2 > (Down > regulated > genes). When I see the fold change value in both tables , > there are > some > genes having fold change value less than 1 in Table 1 and > some genes > have > fold change value greater than 1 in Table 2. > "If the Gene has fold change value less than 1 then its > down regulated" > how > can I interpret fold change value (up regulated or down > regulated ) in > such > case? > > > Theoretically this shouldn't happen as expression level is > suppressed when downregulated. I don't know what cutoff point you > selected for topGene. > If you use a loose criteria, which would lead to gene with not > strong signal being identified out, this would happen. > For example, if gene A has 4 fold-change readout (in one- channel > case) as 1.6,0.9,0.7,0.9 then the average fold-change is1.025. > However, this gene might be identified in downregulation list as 3 > out of 4 fold changes are less than 1. > > > 2) In some cases both Table 1 and Table 2 contains same probe. Is it > > possible to have one probe present in both tables? If Yes, > then which > one > should be considered? > > > This type result would very much indicate this gene doesn't have > decent signal in either direction. This would happen when random > variation gives fake signal in both direction, when > For example, a gene with 4 fold-changes of 1.3, 0.6, 0.9, 1.2 > would results in both list if a loose cut-off point is selected. > > 3) Sometime I see "Inf" as fold change value - must be > infinity. Is it > possible to have such value? > > > If there is 0 or negative value in the expression data (like the > one normalized with MAS5), then it is possible. > > > As stated in package manual, it is always a good to look at the > data when such results coming out, which would help a lot in term > of selecting cut-off point and interpret results > > Hope this would help, let me know if this is not clear or you > prefer to send over your data for mt to take a look. > > Best, > Fangxin > > Could you please help me understanding the above points? > > Thanks and regards, > Rohan > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > <mailto: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 > > > > -- > > Fangxin Hong Ph.D. > Research Scientist > Department of Biostatistics and Computational Biology > Dana-Farber Cancer Institute, Harvard School of Public Health > Phone: 617-632-3602 > Email: fxhong at jimmy.harvard.edu <mailto:fxhong at="" jimmy.harvard.edu=""> > > -- Fangxin Hong Ph.D. Research Scientist Department of Biostatistics and Computational Biology Dana-Farber Cancer Institute, Harvard School of Public Health Phone: 617-632-3602 Email: fxhong at jimmy.harvard.edu