Dear Sean, Thanks for your comments! Are you saying a data set with technical replicates only, like this one, is not appropriate for any limma model, or even regular t-test? This was the concern I had in my first help request. Actually the lab researchers conducted RT-PCR, in which they used two strategies that may improve the uncertainty caused due to lack of biological replicates in microarray assay. One was that they used samples that were from separate mice relative to ones for microarray. Secondly they selected genes with at least 2-fold change in gene expression for PCR verification. The results were pretty consistent between microarray and RT-PCR. Can genes with more than 2-fold change in expression avoid possible dye effect in general? Many thanks again! JP- --On Tuesday, August 01, 2006 11:52 AM -0400 Sean Davis <sdavis2 at="" mail.nih.gov=""> wrote: > > > > On 8/1/06 11:40 AM, "Jianping Jin" <jjin at="" unc.edu=""> wrote: > >> Dear Sean, >> >> Thanks for your reply! I double checked with the lab researcher about the >> sample pooling. As I understood, the total >> RNA was pooled from 3 mice (wt or ko) and then split into 3 aliquots. >> Each aliquot was separately reverse transcripted and labeled. Two >> aliquots of the labeled cDNAs from wt and ko separately were then mixed, >> purified and hybridized onto an Agilent chip. Hope this is clearer. > > So, if I understand correctly, these are not really biologic replicates. > But for the purposes of analysis, they all have the same variance > structure (whatever that is), so can be treated on "equal footing" as far > as analysis is concerned. Since you don't have biological replication, > whatever you find will be of limited biologic generality. In other > words, if one runs the experiment again using different mice, the genes > that you get may be different. > > As I mentioned before, the lack of dye swaps is more problematic, as any > differentially expressed gene (if you find any) will be due to EITHER dye > bias or biologic effect. If you have more than one probe per gene (and > for some genes, that will be the case), and all probes show the same > magnitude and direction of change, that is probably believably not due > entirely to dye effect. However, there is no way to know for sure and > most genes will not have two or more probes that worked for each gene (on > a 44k Agilent array, at least). For "publication" purposes you will > basically have to run dye swaps for such a direct design (unless there is > going to be validation using a second technology such as PCR). > > Of course, there may be other opinions here, and the data can be used for > many purposes besides strictly "publication", so you will need to make up > your own mind in consultation with the lab researcher. > > Sean > >> >> --On Monday, July 31, 2006 2:55 PM -0400 Sean Davis >> <sdavis2 at="" mail.nih.gov=""> wrote: >> >>> >>> >>> >>> On 7/31/06 2:49 PM, "Jianping Jin" <jjin at="" email.unc.edu=""> wrote: >>> >>>> >>>> Dear list: >>>> >>>> There is a data set, consisting of 3 Agilent slides. The experiment was >>>> run with direct hybridization, knock-out versus wild-type, and no dye >>>> swap. Due to difficulty of collecting samples, the samples were pooled >>>> and hybridized onto 3 separate slides. >>> >>> How were the samples pooled? Were they pooled and then split, or are >>> there three distinct biologic replicates? >>> >>> The lack of dye swap IS a problem, as you will likely find dye- biased >>> probes (potentially MANY). >>> >>>> Of course the 3 slides are not biological replicates. They are not pure >>>> technical replicates either. How should I set up a design matrix for >>>> limma model analysis? >>> >>> You'll need to be a bit more specific about how you did the pooling.... >>> >>> Sean >>> >> >> >> >> ################################## >> Jianping Jin Ph.D. >> Bioinformatics scientist >> Center for Bioinformatics >> Room 3133 Bioinformatics building >> CB# 7104, Campus >> Phone: (919)843-6105 >> FAX: (919)843-3103 >> E-Mail: jjin at unc.edu > ################################## Jianping Jin Ph.D. Bioinformatics scientist Center for Bioinformatics Room 3133 Bioinformatics building CB# 7104 University of Chapel Hill Chapel Hill, NC 27599 Phone: (919)843-6105 FAX: (919)843-3103 E-Mail: jjin at email.unc.edu