I have been having an on-going discussion with a colleague about whether he can say that some genes are "absent" in some tissues based on two-color microarrays - most recently, Agilent arrays. There are a number of reasons that he would like to do this which are a mix of biology and QC. He wants to use some (arbitrary) normalized expression level, or unnormalized level above local background or a percentile of the whole array background or ... Any suggestions for papers about this? (We can both think of a dozen ways to do it, but without experiments to see if they are valid methods, or at least a paper to cite, I am reluctant to put the statistical seal of approval on any of them.) Thanks, Naomi p.s. In case anyone thinks that high-throughput sequencing is going to end this type of discussion, have a look at the interesting paper by 't Hoen comparing sequencing and microarray results. http://www.ncbi.nlm.nih.gov/sites/entrez?db=pubmed&cmd=Retrieve&list_u ids=18927111 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

Hi Naomi, I am doing work on AffyChips, but my samples are from amplified material with inputs from around 500pico - few nanogram range; I found that the Affy-Algorithmus "MAS5 calls" was not very happy with this type of data and thus compared expression values on the chip with an empirical negative-distribution (background-distribution), please refer to people.brandeis.edu/~dtaylor/Taylor_Papers/BIBE07_PANP.pdf for details (or the Bioconductor-package panp)... I worked on the Arabidopsis chip, and there are "negative" probes on the chip: probes that do not match DNA sequences from newer genome releases anymore... I found that by combining the panp-strategy and the information on negative probes on the chip, I could generate precise and relatively accurate predictions on the expression state of a gene (I can't give you all the details so far). So if there are negative controls/negative probes on the array, you could use them to generate an empirical background-distribution for each array and then compare your other signals to this.... Depends on how many negative probes you'd have... The method works well with the Affy HGU133-series, please refer to the above mentioned sources... Hope this helps?? Best, Sam 2009/1/30 Naomi Altman <naomi at="" stat.psu.edu="">: > I have been having an on-going discussion with a colleague about whether he > can say that some genes are "absent" in some tissues based on two- color > microarrays - most recently, Agilent arrays. There are a number of reasons > that he would like to do this which are a mix of biology and QC. > > He wants to use some (arbitrary) normalized expression level, or > unnormalized level above local background or a percentile of the whole array > background or ... > > Any suggestions for papers about this? (We can both think of a dozen ways > to do it, but without experiments to see if they are valid methods, or at > least a paper to > cite, I am reluctant to put the statistical seal of approval on any of > them.) > > Thanks, Naomi > > p.s. In case anyone thinks that high-throughput sequencing is going to end > this type of discussion, have a look at the interesting paper by 't Hoen > comparing sequencing and microarray results. > http://www.ncbi.nlm.nih.gov/sites/entrez?db=pubmed&cmd=Retrieve&list _uids=18927111 > > 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 > > _______________________________________________ > 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 > >