Dear Anand, It seems to me the simplest thing would be to consider patient as the block and viral titre as either a continuous covariate. If the slope differs among the strains, you will need to include an interaction term. --Naomi At 02:51 PM 2/28/2011, Anand Patel wrote: >I'm struggling with the best design for modeling effects of different >viral strains in a complex experiment. > >Factors: >1) Patient (p3, p4, p5) >2) "Replicate" (a, b, c) >3) Viral Titer (continuous integer variable) >4) Viral Strain (O, F, S) > >Although all 3 of the "replicates" per patient were treated the same >way, there are significant differences in the amount of virus >recovered from each "replicate", and that appears to have a >significant effect on gene expression (based on multivariate >projection mapping plots). As this is a biologically plausible >result, I'm trying to figure out a way to include the titer >information in a model while not treating the "replicates" as fully >independent. > >This is complicated by the 0 titer occurring only in the untreated >wells (again, this makes sense, but makes modeling a challenge). > >Using duplicateCorrelation without regards to the experimental design, >I get a corfit$cor of 0.3790526 . > >When I use duplicateCorrelation using: >design <- model.matrix(~0+p+v) >(where p and v are factors representing patient and viral strain, >respectively) > >I get a corfit$cor of 0.1430260. > >While titer is related to the individual patient, it's acting >independently based on mds plots of individual patient gene >expression, but I'm just not sure how to best model this experiment. > >Thoughts? > >Thanks, >Anand > >_______________________________________________ >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