Matt, does probeVec represents some sort of baseline expression for each specific probe? does it make sense to calculate these vectors for another affy chip with 20 samples? thanks! robert. On 3/5/12 7:52 PM, Matthew McCall wrote: > Robert, > > I'm not sure exactly what you're after, but you might want to look at > the hgu133afrmavecs and hgu133plus2frmavecs data packages. The > probe-effect (probeVec), within-batch residual variance > (probeVarWithin), the between-batch residual variance > (probeVarBetween), and the within probeset standard deviation > (probesetSD) have all been computed using a large biologically diverse > data set. > > Best, > Matt > > On Mon, Mar 5, 2012 at 1:22 PM, Robert Castelo<robert.castelo at="" upf.edu=""> wrote: >> dear list, >> >> i'm searching for a way to rank affy probesets from classical 3' affy >> arrays by their probe effect magnitude. i mean that i would like to know >> if a probeset is has a larger probe-specific effect than another one. >> >> i guess the solution should be in the affyPLM package since if i do >> >> library(affy) >> library(affyPLM) >> >> ab<- ReadAffy() >> pset<- fitPLM(ab) >> >> >> i obtain an object (pset) of the PLMset class which contains slots >> 'probe.coefs' and 'se.probe.coefs', where each is a list as many keys as >> probesets and where each probeset contains information on the probe >> effect of each probe within the probeset: >> >> head(names(pset at probe.coefs)) >> [1] "1000_at" "1001_at" "1002_f_at" "1003_s_at" "1004_at" >> "1005_at" >> head(names(pset at se.probe.coefs)) >> [1] "1000_at" "1001_at" "1002_f_at" "1003_s_at" "1004_at" >> "1005_at" >> >> pset at probe.coefs[[1]] >> Overall >> probe_1 0.97287528 >> probe_2 0.61454806 >> probe_3 -2.81701693 >> probe_4 1.68063395 >> probe_5 -3.31991235 >> probe_6 1.56657388 >> probe_7 -3.30256264 >> probe_8 -1.99431231 >> probe_9 -0.35200585 >> probe_10 -0.49024387 >> probe_11 -1.09087811 >> probe_12 0.22008832 >> probe_13 2.54263342 >> probe_14 3.71106614 >> probe_15 2.12580554 >> probe_16 -0.06729251 >> >> pset at se.probe.coefs[[1]] >> Overall >> probe_1 0.06124122 >> probe_2 0.06039453 >> probe_3 0.06180433 >> probe_4 0.05948503 >> probe_5 0.06727454 >> probe_6 0.06016827 >> probe_7 0.06233682 >> probe_8 0.06791376 >> probe_9 0.05960599 >> probe_10 0.05963511 >> probe_11 0.05868359 >> probe_12 0.06046023 >> probe_13 0.05885199 >> probe_14 0.05829506 >> probe_15 0.05837877 >> probe_16 0.06340662 >> >> however, i'm unsure how to proceed from now on to decide whether a >> particular probeset is more "affected" by probe-specific effects than >> other probeset. any suggestion would be highly appreciated, >> >> thanks, >> robert. >> >> _______________________________________________ >> 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 > > >

hi Jim, On 3/5/12 8:02 PM, James W. MacDonald wrote: [...] > I'm not sure what you are looking to do with these data, but remember > that the probe-specific effects in RMA are estimated as a nuisance > variable, which are then excluded from computation of the expression > value. So by definition the probesets should not be affected by > probe-specific effects. my understanding is that probesets in a microarray may have different baseline expression levels due to probe-specific effects. if i recall correctly this is illustrated in Fig. 1 of Zilliox and Irizarry (Nat. Meth., 2007) and this, for instance, complicates the question of determining whether a gene is expressed or not, which is tackled on that paper. i would like to assess somehow the agreement of these possibly different expression baselines between different genes and i thought that probe-specific effects would contain such information. to be more specific, if we would assume a simple additive model y_ij = probe_effect_i + sample_effect_j + noise_ij i'd like to know what genes have more similar or more disparate probe effects estimated as the probe_effect_i term in the previous linear model and i thought such information could be extracted from PLMset objects. cheers, robert.