Hi, Hopefully someone can help me with this one. I'm fairly sure what might be happening but would like confirmation before I pass judgement to the experimenter. This is basically linked to my earlier post where I was looking at scatterplots to assess reproducibility between replicates. Anyway, I've move backwards towards the raw data, and I think I have an idea of what is happening. I've attached some files again, they're scrubbed from the email, but you can find them on the html version of the list https://www.stat.math.ethz.ch/pipermail/bioconductor/2004-July/date.ht ml The first is the density plot of the Log'd raw intensities. It shows 6 chips. All the red channel intensities peak around 12, the green at 15. All the green channels have a low peak with an extended shoulder to the right. chips 1-3 of the red channel have a similar pattern to the green, whilst 4-6 have more of a spike at 12 and less of a shoulder (4-6 look more like the example in the limma guide!). If you do a MA plot of any of chips 1-3 (+/- BG correct x +/- normalisewithinarrays), then the MA plot has a 'finger' coming out upwards (5? from horizontal) (see 2nd attach). The BG images of at least 2 of the 3 chips look fine). I think this could be a labelling problem, but still have some questions. Why is is only at higher intensities? Is the red the bad label, or (if unlabelled DNA is removed? and taking into account the competitive mixing?) is it actually the green label that is bad? (I just look at the data) But how do people usually QC the labelling and ensure that exactly the same amount of the 2 samples are mixed? Thanks in advance, Matt -------------- next part -------------- A non-text attachment was scrubbed... Name: Raw intensities.png Type: image/png Size: 8628 bytes Desc: Raw intensities.png Url : https://www.stat.math.ethz.ch/pipermail/bioconductor/attachments /20040729/34b1cb96/Rawintensities.png -------------- next part -------------- A non-text attachment was scrubbed... Name: e95 MA no BG.png Type: image/png Size: 9683 bytes Desc: e95 MA no BG.png Url : https://www.stat.math.ethz.ch/pipermail/bioconductor/attachments /20040729/34b1cb96/e95MAnoBG.png

Matt, I find it useful to look at something like maImage plots (looking at array values as if they were on the array) to examine for spacial artifacts. You need to specify your array layout to do this (and the functionality is available via Limma or marray). This could be something like a local background effect (washing all arrays in one direction, for example) or a pin failure on the print used for these experiments. Sometimes it is obvious at the point of image extraction, but sometimes such effects are only observable after some image analysis and information summary. Ratios should correct somewhat any differences in amount of loaded RNA, so to have just a subset of points being "outliers" is not easily explained by such a systematic error. Sean On Jul 29, 2004, at 10:52 AM, Matthew Hannah wrote: > Hi, > > Hopefully someone can help me with this one. I'm fairly sure what > might be happening but would like confirmation before I pass judgement > to the experimenter. This is basically linked to my earlier post where > I was looking at scatterplots to assess reproducibility between > replicates. > > Anyway, I've move backwards towards the raw data, and I think I have > an idea of what is happening. I've attached some files again, they're > scrubbed from the email, but you can find them on the html version of > the list > https://www.stat.math.ethz.ch/pipermail/bioconductor/2004-July/ > date.html > > The first is the density plot of the Log'd raw intensities. It shows 6 > chips. All the red channel intensities peak around 12, the green at > 15. All the green channels have a low peak with an extended shoulder > to the right. chips 1-3 of the red channel have a similar pattern to > the green, whilst 4-6 have more of a spike at 12 and less of a > shoulder (4-6 look more like the example in the limma guide!). > > If you do a MA plot of any of chips 1-3 (+/- BG correct x +/- > normalisewithinarrays), then the MA plot has a 'finger' coming out > upwards (5? from horizontal) (see 2nd attach). The BG images of at > least 2 of the 3 chips look fine). I think this could be a labelling > problem, but still have some questions. > Why is is only at higher intensities? > Is the red the bad label, or (if unlabelled DNA is removed? and taking > into account the competitive mixing?) is it actually the green label > that is bad? > (I just look at the data) But how do people usually QC the labelling > and ensure that exactly the same amount of the 2 samples are mixed? > > Thanks in advance, > Matt > > <raw intensities.png=""><e95 ma="" no=""> BG.png>_______________________________________________ > Bioconductor mailing list > Bioconductor@stat.math.ethz.ch > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor

Sean, Thanks for the swift reply! I've used these already - >imageplot(log2(RG$Rb[,3]), RG$printer, low="white", high="red") #also for Gb and all chips No consistent BG effects >imageplot(MA.nb$M[,3], RG$printer, zlim=c(-3,3)) #this was on no BG correction, normalisewithinarrays Alot of bright red spots show up across the whole array, but no consistent patterns across all 3. Since your mail, I've just tried the second without normalisewithinarrays (ie: method="none"), obviously the array is mostly green as the channels have not been scaled, but you can still see alot of the red points showing through, but again no patterns. So it leaves the fact that I have a bias for some (a lot) probes towards the red channel, particularly at high intensities. Other ideas? Thanks, Matt -----Original Message----- From: Sean Davis [mailto:sdavis2@mail.nih.gov] Sent: Donnerstag, 29. Juli 2004 17:12 To: Matthew Hannah Cc: bioconductor@stat.math.ethz.ch Subject: Re: [BioC] 2-colour bias AFTER normalising with Limma, labelling QC Matt, I find it useful to look at something like maImage plots (looking at array values as if they were on the array) to examine for spacial artifacts. You need to specify your array layout to do this (and the functionality is available via Limma or marray). This could be something like a local background effect (washing all arrays in one direction, for example) or a pin failure on the print used for these experiments. Sometimes it is obvious at the point of image extraction, but sometimes such effects are only observable after some image analysis and information summary. Ratios should correct somewhat any differences in amount of loaded RNA, so to have just a subset of points being "outliers" is not easily explained by such a systematic error. Sean On Jul 29, 2004, at 10:52 AM, Matthew Hannah wrote: > Hi, > > Hopefully someone can help me with this one. I'm fairly sure what > might be happening but would like confirmation before I pass judgement > to the experimenter. This is basically linked to my earlier post where > I was looking at scatterplots to assess reproducibility between > replicates. > > Anyway, I've move backwards towards the raw data, and I think I have > an idea of what is happening. I've attached some files again, they're > scrubbed from the email, but you can find them on the html version of > the list > https://www.stat.math.ethz.ch/pipermail/bioconductor/2004-July/ > date.html > > The first is the density plot of the Log'd raw intensities. It shows 6 > chips. All the red channel intensities peak around 12, the green at > 15. All the green channels have a low peak with an extended shoulder > to the right. chips 1-3 of the red channel have a similar pattern to > the green, whilst 4-6 have more of a spike at 12 and less of a > shoulder (4-6 look more like the example in the limma guide!). > > If you do a MA plot of any of chips 1-3 (+/- BG correct x +/- > normalisewithinarrays), then the MA plot has a 'finger' coming out > upwards (5? from horizontal) (see 2nd attach). The BG images of at > least 2 of the 3 chips look fine). I think this could be a labelling > problem, but still have some questions. > Why is is only at higher intensities? > Is the red the bad label, or (if unlabelled DNA is removed? and taking > into account the competitive mixing?) is it actually the green label > that is bad? > (I just look at the data) But how do people usually QC the labelling > and ensure that exactly the same amount of the 2 samples are mixed? > > Thanks in advance, > Matt > > <raw intensities.png=""><e95 ma="" no=""> BG.png>_______________________________________________ > Bioconductor mailing list > Bioconductor@stat.math.ethz.ch > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor