It is adhoc and I?ve only thought a little about how to do regional correction more systematically (the way field studies correct for wet/soil spots locally effecting growth). Something like the xy pos of the feature is in the model for differential expression. This would be quite computational and normalization wouldn?t come in the standard way either. So bg.correct from RMA is not spatial correction. It doesn?t consider the XY pos. It does a global per array correction for each array, but assumes normal noise and exponential RNA signal. Ben Bolstad was going to model normal noise and normal signal for DNA, but I don?t think he has done it yet, no pressure Ben... We might be in lower demand however SNP arrays might benefit tremendously from normal/normal. SNP arrays do have their own correction method with the oligo package though and I?m not sure if anyone tired this for RNA or if it makes sense even. Yes quantile normalization is the next step and good luck with the maize SFPs, if you use whole genome random labeling I guess it will be noisily but possible with enough replicates. The deletions sure should be. Otherwise a complexity reduction eg AFPL etc might help, see our Barley paper http://genomebiology.com/2005/6/6/r54 ----- Justin Borevitz http://naturalsystems.org/lab ________________________________________ From: Michael Gore [mailto:mag87@cornell.edu] Sent: Friday, October 13, 2006 6:08 PM To: 'Justin Borevitz' Subject: Spatial Correction Hi Justin, ? I have a quick question for you.? How does one determine the appropriate array size and filter size for spatial correction? ? Is spatial correction comparable to RMA, followed by Quantiles? ? Right now, we are doing SFP with the Affy Maize GeneChip. ? Thanks, ? Mike ? ? library(affy) read.cel <- function(cel.file, cel.image = F, spatial.correct=T, median=F, ?????????????????????????????????? array.size = 712, filter.size = 51, jpeg.save=T){ ?

heh. we're often on the same wavelength it seems. i'm already doing this x-y slope across the whole 'field' thing a bit in my linear modeling work, inspired by borevitz's occosional inclusion of probe set id #'s as null- components when checking stats/models, instead i input them as x and y's, split up as terms in a model for calling the chip. i have to say: the coeffiecients sure don't look like they matter much? i think i'm going to leave it in as an engineering/q.a. thing, though. e.g. coeffiecients for a model that slams everything down to snp/sfp-calls/copy-number (sorry no anova, sigs, or f-test output on these runs): statcutoffs tcut: 12.236 high: 21.2475972466423 low: 16.5098342910872 built model of 42317 points, spread as transects across the chip. constant 2.06606598216225 --- progenypm 0.0813258560574814 progenymm -0.00294383026705664 highpm1 -0.0486924459388655 highpm2 -0.0365650706946507 highpm3 -0.046313242027037 lowpm1 0.0181522513255337 lowpm2 0.0161600597386146 lowpm3 0.0160355165311527 highmm1 -0.0110724545083531 highmm2 -0.00663396018469053 highmm3 -0.0116972092671024 lowmm1 0.00326259465264771 lowmm2 0.00567155215296054 lowmm3 0.0013044918026059 -- gc content: gcp (sign gets flipped) -0.00101239058730091 -- nearest neighbor temperature: tmp (sign gets flipped) 0.00153804487609968 -- number of G's in probe (the c's get labeled): ngp (sign gets flipped) 0.00554060155127831 --- xpos (sign gets flipped) -2.14773325615418e-05 <-- x 10 to the minus 5, that's not-so-much? ypos (sign gets flipped) 6.94183011922954e-06 --- as far as abberant spot correction (real spatial correction by-zones) i'm not doing this and am simply re-running the bad experiements - for some reason a few of our chips have smudges that look like nike-swooths on them where the hybridization is killed significantly. --- as far as labeling for dna is concerned i would die to see someone compare end-transferase and random label inclusion (perhaps in the same hyb mix as different fluorophore colors on a multi-channel scanner?). is there any work / pubs out there that do this? Thank You, Matthew Lyon UC Riverside lab (951) 827-4736 Ph.D. Student B O T A N Y new c.p. (951) 941-5554 Citrus Genomics apt (951) 328-9930 http: // int - citrusgenomics . org / messengers: ptrifoliata mattlyon at mattlyon.com ptrifoliata at hotmail.com mlyon003 at student.ucr.edu >From: "Justin Borevitz" <borevitz at="" uchicago.edu=""> >To: "'Michael Gore'" <mag87 at="" cornell.edu=""> >CC: Ben Bolstad <bolstad at="" stat.berkeley.edu="">, "'Bioconductor'" ><bioconductor at="" stat.math.ethz.ch=""> >Subject: Re: [BioC] Spatial Correction >Date: Sat, 14 Oct 2006 09:53:52 -0500 > >It is adhoc and I?ve only thought a little about how to do regional >correction more systematically (the way field studies correct for wet/soil >spots locally effecting growth). Something like the xy pos of the feature >is in the model for differential expression. This would be quite >computational and normalization wouldn?t come in the standard way either. > >So bg.correct from RMA is not spatial correction. It doesn?t consider the >XY >pos. It does a global per array correction for each array, but assumes >normal noise and exponential RNA signal. Ben Bolstad was going to model >normal noise and normal signal for DNA, but I don?t think he has done it >yet, no pressure Ben... We might be in lower demand however SNP arrays >might benefit tremendously from normal/normal. SNP arrays do have their own >correction method with the oligo package though and I?m not sure if anyone >tired this for RNA or if it makes sense even. > >Yes quantile normalization is the next step and good luck with the maize >SFPs, if you use whole genome random labeling I guess it will be noisily >but >possible with enough replicates. The deletions sure should be. Otherwise >a >complexity reduction eg AFPL etc might help, see our Barley paper >http://genomebiology.com/2005/6/6/r54 > > >----- >Justin Borevitz >http://naturalsystems.org/lab >________________________________________ >From: Michael Gore [mailto:mag87 at cornell.edu] >Sent: Friday, October 13, 2006 6:08 PM >To: 'Justin Borevitz' >Subject: Spatial Correction > >Hi Justin, > >I have a quick question for you. How does one determine the appropriate >array size and filter size for spatial correction? > >Is spatial correction comparable to RMA, followed by Quantiles? > >Right now, we are doing SFP with the Affy Maize GeneChip. > >Thanks, > >Mike > > >library(affy) >read.cel <- function(cel.file, cel.image = F, spatial.correct=T, median=F, > array.size = 712, filter.size = 51, >jpeg.save=T){ > > >_______________________________________________ >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

Cool can u give us the form of the linear model? ----- Justin Borevitz http://naturalsystems.org/lab --> sure. it's a multi-term model (engineering-kludge), where degrees of freedom suffer and statistical validity is not the goal, just polishing the data is. it?s designed to call a single chip?d progeny with respect to the parents, run per probe: y_call= B0 + B1*X_progenypm + B2*X_progenymm + B3*X_highpm1 + B4*X_highpm2 + B5*X_highpm3 + B6*X_lowpm1 + B7*X_lowpm2 + B8*X_lowpm3 + B9*X_highmm1 + B10*X_highmm2 + B11*X_highmm3 + B12*X_lowmm1 + B13*X_lowmm2 + B14*X_lowmm3 - B15*X_gcp - B16*X_tmp - B17*X_ngp - B18*X_xpos < - these are the two you?re interested in. - B19*X_ypos < - o where y is a borevitizian sfp-call using a non-permed t-test, empirically set to force ?polymorphisms? on about 10% of the chip (oversplitting) o where the perfect match probes, mismatch probes, two parents, and progeny are in the named terms, with the parents always ordered high-to-low (hence, high, low) o where the gcp, tmp, ngp, xpos, and ypos are build with + in the sign, then the sign is flipped at evaluation time. o where the output is then split empirically, as well. Thank You, Matthew Lyon UC Riverside lab (951) 827-4736 Ph.D. Student B O T A N Y new c.p. (951) 941-5554 Citrus Genomics apt (951) 328-9930 http: // int - citrusgenomics . org / messengers: ptrifoliata mattlyon at mattlyon.com ptrifoliata at hotmail.com mlyon003 at student.ucr.edu