Question: Low Ratio values in LIMMA
0
gravatar for Nataliya Yeremenko
14.0 years ago by
Nataliya Yeremenko100 wrote:
Hello everybody After successful importing the data to the LIMMA, I followed all steps and finally obtained "topTable". (the data were 44K Agilent) I'm surprised that my data set doesn't have anything with M-values more than 1.5. I well understand that it depends on the experiment, but is there some condensation of the Ratio during normalization and other procedures? regards -- Dr. Nataliya Yeremenko Universiteit van Amsterdam Faculty of Science IBED/AMB (Aquatische Microbiologie) Nieuwe Achtergracht 127 NL-1018WS Amsterdam the Netherlands tel. + 31 20 5257089 fax + 31 20 5257064
normalization limma • 520 views
ADD COMMENTlink modified 14.0 years ago by Sean Davis21k • written 14.0 years ago by Nataliya Yeremenko100
Answer: Low Ratio values in LIMMA
0
gravatar for Greg Finak
14.0 years ago by
Greg Finak50
Greg Finak50 wrote:
Hi, Nataliya. This is very much dependent on how you normalize your data. It's possible to "over normalize" and wipe out most of the variability in your data. If you're certain that this isn't the case, then, no, Limma doesn't generate shrunken estimates of the expression ratios, only t-statistics. The expression ratios presented by limma will either be log2, or a data dependent scale if you've used VSN to normalize. Feel free to post more information about your normalization / preprocessing / experimental design, that would help to identify potential reasons, otherwise we're just speculating. Cheers, -------------------------------------------- Greg Finak PhD Candidate McGill Center For Bioinformatics W: (514)398-7071 x09317 emai: finak at mcb.mcgill.ca -------------------------------------------- On 9-Nov-05, at 7:13 PM, Nataliya Yeremenko wrote: > Hello everybody > > After successful importing the data to the LIMMA, > I followed all steps and finally obtained "topTable". > (the data were 44K Agilent) > I'm surprised that my data set doesn't have anything with M-values > more > than 1.5. > I well understand that it depends on the experiment, > but is there some condensation of the Ratio during normalization and > other procedures? > > regards > > -- > Dr. Nataliya Yeremenko > > Universiteit van Amsterdam > Faculty of Science > IBED/AMB (Aquatische Microbiologie) > Nieuwe Achtergracht 127 > NL-1018WS Amsterdam > the Netherlands > > tel. + 31 20 5257089 > fax + 31 20 5257064 > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor
ADD COMMENTlink written 14.0 years ago by Greg Finak50
Dear Greg Thanks for your comment: Detailed description and a number of problems are described in my new BioC post with subject: Design question in LIMMA As for data preporcessing step I used ordinary: RG <- backgroundCorrect(RG,method="minimum") MA <-normalizeWithinArrays(RG, method="loess") MA <- normalizeBetweenArrays(MA, method="Aquantile") I didn't use weighting of the control spots (commercial Agilent 44K) and didn't used anything flaged by Feature Extraction. Maybe I have to? I visualize the data afterwards and they look not bad. I do understand that the data in topTable are in log2 scale, but the differences seem to me to be really small. The problem may come as well from the wrong design and contrasts created. I hope this makes clear procedures I used. Regards Nataliya Greg Finak wrote: > Hi, Nataliya. > > This is very much dependent on how you normalize your data. It's > possible to "over normalize" and wipe out most of the variability in > your data. If you're certain that this isn't the case, then, no, > Limma doesn't generate shrunken estimates of the expression ratios, > only t-statistics. The expression ratios presented by limma will > either be log2, or a data dependent scale if you've used VSN to > normalize. Feel free to post more information about your > normalization / preprocessing / experimental design, that would help > to identify potential reasons, otherwise we're just speculating. > > Cheers, > > > -------------------------------------------- > Greg Finak > PhD Candidate > McGill Center For Bioinformatics > > W: (514)398-7071 x09317 > emai: finak at mcb.mcgill.ca > -------------------------------------------- > > On 9-Nov-05, at 7:13 PM, Nataliya Yeremenko wrote: > >> Hello everybody >> >> After successful importing the data to the LIMMA, >> I followed all steps and finally obtained "topTable". >> (the data were 44K Agilent) >> I'm surprised that my data set doesn't have anything with M-values more >> than 1.5. >> I well understand that it depends on the experiment, >> but is there some condensation of the Ratio during normalization and >> other procedures? >> >> regards >> >> -- >> Dr. Nataliya Yeremenko >> >> Universiteit van Amsterdam >> Faculty of Science >> IBED/AMB (Aquatische Microbiologie) >> Nieuwe Achtergracht 127 >> NL-1018WS Amsterdam >> the Netherlands >> >> tel. + 31 20 5257089 >> fax + 31 20 5257064 >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor at stat.math.ethz.ch >> https://stat.ethz.ch/mailman/listinfo/bioconductor > > > > -- Dr. Nataliya Yeremenko Universiteit van Amsterdam Faculty of Science IBED/AMB (Aquatische Microbiologie) Nieuwe Achtergracht 127 NL-1018WS Amsterdam the Netherlands tel. + 31 20 5257089 fax + 31 20 5257064
ADD REPLYlink written 14.0 years ago by Nataliya Yeremenko100
Answer: Low Ratio values in LIMMA
0
gravatar for Sean Davis
14.0 years ago by
Sean Davis21k
United States
Sean Davis21k wrote:
On 11/9/05 7:13 PM, "Nataliya Yeremenko" <eremenko at="" science.uva.nl=""> wrote: > Hello everybody > > After successful importing the data to the LIMMA, > I followed all steps and finally obtained "topTable". > (the data were 44K Agilent) > I'm surprised that my data set doesn't have anything with M-values more > than 1.5. > I well understand that it depends on the experiment, > but is there some condensation of the Ratio during normalization and > other procedures? > > regards Nataliya, If I understand your original question, you are asking about M-values as reported by topTable? These M-values are the log2 fold-change associated with the contrast associated with that coefficient. In other words, if you are doing a two-group comparison in limma (as a simple example), the M-value represents the average log2 fold-change between the groups. If you are not getting large M-values, then you do not have large differences between your groups. Of course, normalization can have a significant effect on the fold- changes that you see in topTable, but another likely explanation, normalization aside, is that there are not large differences between groups. I may have misunderstood your question, but I'm not sure that normalization is the culprit here. And I think that most would advocate doing normalization based on biologic understanding of the experiment and global characteristics of the data and not the resulting fold-changes or significance; it really probably isn't a good idea to do multiple normalizations looking for the one that gives the best fold-change or gene list (not that you were going to do so). Hope this helps, Sean
ADD COMMENTlink written 14.0 years ago by Sean Davis21k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 381 users visited in the last hour