Dear All, say I have the following samples: f1_A1 f1_A2 f1_B1 f1_B2 f2_C1 f2_C2 f2_D1 f2_D2 Here f1 and f2 are two families A1 and A2 are technical replicate. same for B1 and B2,C1 and C2, D1 and D2. A,B,C,D are biological replicate, say A,C are affected while B,D are unaffected. Noted some families have more than 1 affected or unaffected samples. How does LIMMA take the two-level (technical replicate and familial relationship) dependence into account? Can I just simply do this way: When I calculate duplicateCorrelation, I can set my block as myblock<-c(1,1,2,2,3,3,4,4), #here take the technical replicate into account When I do lmFit, I set the design and block as s<- factor(c(1,1,0,0,1,1,0,0)) #1-affected and 0-unaffected design <- model.matrix(~0+s) myblock<-c(1,1,1,1,2,2,2,2) #two-families Here I used two different blocks in duplicateCorrelation and lmFit, looks not right.... Thanks, Qin Kuang _________________________________________________________________ [[alternative HTML version deleted]]

Dear All, say I have the following samples: f1_A1 f1_A2 f1_B1 f1_B2 f2_C1 f2_C2 f2_D1 f2_D2 Here f1 and f2 are two families A1 and A2 are technical replicate. same for B1 and B2,C1 and C2, D1 and D2. A,B,C,D are biological replicate, say A,C are affected while B,D are unaffected. Noted some families have more than 1 affected or unaffected samples. How does LIMMA take the two-level (technical replicate and familial relationship) dependence into account? Can I just simply do this way: When I calculate duplicateCorrelation, I can set my block as myblock<-c(1,1,2,2,3,3,4,4), #here take the technical replicate into account When I do lmFit, I set the design and block as s<- factor(c(1,1,0,0,1,1,0,0)) #1-affected and 0-unaffected design <- model.matrix(~0+s) myblock<-c(1,1,1,1,2,2,2,2) #two-families Here I used two different blocks in duplicateCorrelation and lmFit, looks not right.... Thanks, Qin Kuang _________________________________________________________________ [[alternative HTML version deleted]]

Dear All, Anyone can kindly provide me some insights/comments on how to analyze the data using either LIMMA or other tools? Thanks, Qin -------------------------------------------------- Dear All, say I have the following samples: f1_A1 f1_A2 f1_B1 f1_B2 f2_C1 f2_C2 f2_D1 f2_D2 Here f1 and f2 are two families A1 and A2 are technical replicate. same for B1 and B2,C1 and C2, D1 and D2. A,B,C,D are biological replicate, say A,C are affected while B,D are unaffected. Noted some families have more than 1 affected or unaffected samples. How does LIMMA take the two-level (technical replicate and familial relationship) dependence into account? Can I just simply do this way: When I calculate duplicateCorrelation, I can set my block as myblock<-c(1,1,2,2,3,3,4,4), #here take the technical replicate into account When I do lmFit, I set the design and block as s<- factor(c(1,1,0,0,1,1,0,0)) #1-affected and 0-unaffected design <- model.matrix(~0+s) myblock<-c(1,1,1,1,2,2,2,2) #two-families Here I used two different blocks in duplicateCorrelation and lmFit, looks not right.... Thanks, Qin Kuang _________________________________________________________________ [[alternative HTML version deleted]]

If you want to take the family correlation structure into account, limma is not the appropriate software. MAANOVA can handle balanced random effects type data - this would use a random effect that induces equal correlation among all family members regardless of degree of relatedness. To handle unbalanced family data or familial effects determined by the degree of relatedness, you need the more flexible correlation structure afforded by software such as lme. However, this does not give you the added power afforded by (empirical) Bayesian shrinkage. As well, you would need to fit gene by gene, which will definitely mean very slow run time. --Naomi At 02:19 PM 5/13/2010, Steve Lianoglou wrote: >Hi Qin, > >On Thu, May 13, 2010 at 11:38 AM, qin kuang <kuang_qin at="" hotmail.com=""> wrote: > > Thanks Steve. > > > > Actually I have done this family by family. I also tried method like GEE to > > take familial relationship by averaging technical replicates. At > this point, > > looks the methods like LIMMA/GEE can not two-level/type of > dependence (like > > technical replicate and familial relationship) but only one- level/type of > > dependence. > > > > The reason that I want to look all families together is that some > > individuals in these families share the same deletion region > > (this is mRNA expression data, the deletion region was detected in other > > genomic study). In my previous post, I used 'affected' vs 'unaffected'. > > Precisely, it is 'deleted' vs 'undeleted'. > > > > In your point 2, what do you mean "combine families as biological > replicates > > and do cases vs. controls"? This is what I want to do. As I > mentioned above, > > I can only consider one type of dependence, but the data has two types of > > dependence. > >If that's the case, the limma user's guide goes into a good amount of >detail about being careful with replicate data: biological and >technical replicates, and how to rig up your code to incorporate both. > >See section 8.2, for instance. > >Is this what you're after? > >-- >Steve Lianoglou >Graduate Student: Computational Systems Biology > | Memorial Sloan-Kettering Cancer Center > | Weill Medical College of Cornell University >Contact Info: http://cbio.mskcc.org/~lianos/contact > >_______________________________________________ >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 Naomi S. Altman 814-865-3791 (voice) Associate Professor Dept. of Statistics 814-863-7114 (fax) Penn State University 814-865-1348 (Statistics) University Park, PA 16802-2111