I tried this approach, but I believe that the problem is that spots that are adjacent on the array are expected to have higher correlation than spots that are in the upper and lower half of the array for instance. This means that duplicates that are lets say 10 spots apart probably are more correlated than those 200 spots apart, and I am afraid may "disturb" the analysis. I maybe have made some mistakes during the programming (I''m an R beginner), but I have calculated the correlation between duplicate spots in this way for different methods of background correction and normalisation. I thought that the methods giving the highest correlation would be best for further analyses, but the highest correlation was obtained with no background correction and no normalisation. I found this very strange since the background is not uniform within the arrays, and all litterature says that microarray data should be normalised. Ingunn ----- Original Message ----- From: "Jason Skelton" <jps@sanger.ac.uk> To: <ingunn.berget@umb.no>; <bioconductor@stat.math.ethz.ch> Sent: Friday, January 07, 2005 1:01 PM Subject: random location of duplicate spots and use of limma > > >> >>Hello >> >>There are approximately 6000 different genes on the arrays, there are two >>spots for each gene >>The duplicated spots have random location, which means that the number of >>spots between each duplicate is not the same for every gene. This is the >>summary for the distances: >> >> Min. 1st Qu. Median Mean 3rd Qu. Max. 4.00 32.00 71.00 >> 86.59 135.00 244.00 >>(Distance here means number of spots between the two duplicates) >> >>The function duplicateCorrelation in limma can be used to estimate >>correlation between within-array duplicates, the methodology is based on >>the assumption that duplicates are equally spaced. Since this assumption >>is not fulfilled here does this means that I cannot calculate the >>correlations and must take the average of the duplicates? Are there some >>functions to do this in limma or other BioC packages >> > > Hi ingunn & all > > I could be wrong about this but can you get round this in limma by: > normalising the data first(to allow for the physical location on the > array) > followed by re-arranging the normalised data so that duplicate genes > appear next to each other > and therefore have equal spacing ? I.e spacing of 1 or similar. > You obviously have to make new genelists for the "rearranged" order but I > can't see any obvious problems with > further analysis such as the linear model fitting etc. If you only have > two replicates then it should be ok...... > I do this routinely but the limma authors might be able to suggest a > better alternative ? > > > Jason > > > > > > > >>-- >>-------------------------------- >>Jason Skelton >>Pathogen Microarrays >>Wellcome Trust Sanger Institute >>Hinxton >>Cambridge >>CB10 1SA >> >>Tel +44(0)1223 834244 Ext 7123 >>Fax +44(0)1223 494919 >>-------------------------------- >> >

From: "Ingunn Berget" <ingunn.berget@umb.no> I thought that the methods giving the highest correlation would be best for further analyses, but the highest correlation was obtained with no background correction and no normalisation. I found this very strange since the background is not uniform within the arrays, and all litterature says that microarray data should be normalised. Ingunn -------- Ingunn: Please check the intensity-level of the spots. If they are mostly low (that is spot-intensity is close to background-intensity), then background-correction would introduce greater variability among spot-replicates. Something of this sort is reported by Speed'S group in the context of microarray image analysis. Yang et al (2000) Comparison of methods for image analysis on cDNA microarray data. Technical report # 584,Nov 2000. ------ Saroj K Mohapatra, MD Research Associate Wayne State University School of Medicine 110 E Warren Avenue, # 311 Detroit MI 48201 saroj@wayne.edu