See below: On Jan 21, 2013, at 4:36 PM, hu duan wrote: Hi Evan, Thanks for your quick response. I may list some interesting result. 1. The density plot on the left bottom in previous email shown red line had a sharp peak while black not. Can you inform me what does that difference mean? Black = density estimate of the batch effects (used directly by non- parametric adjustments) Red= Normal/invgamma distributions used for the parametric adjustments Ultimately, if the black curve in the first plot is roughly bell shaped, you won't see much difference in the batch effect adjustments. 1. I just finished analysis of the nonparametric and parametric for a same data. The following two plots shown here, which the Y axis is the parametric and X axis is the nonparametric. It seems nonparametric approach will tend to increase signal. The mean for CV for that is 1.7% while median is 0.37%. The data looks pretty close. Yes, they are usually very similar unless the black and red lines are very different from each other. 1. 2. For the run time, I did a PCA analysis by using all features. The left pic shown different run times in different color, which clearly shown run time difference. Right one is after adjust by parametric approach. Would you consider batch effects have been removed? Yes, looks like the batch effects are gone. 1. 2. If I run Combat analysis in same data and use the adjust data to run Combat again, do you think the result will be better or it will overfitting your model? Does the model converge? Yes, you might be overfitting a little, but the EB helps reduce the amount of over-fitting. Also, due to you experimental design, over- fitting shouldn't be a huge issue. 1. Best Tiger <image.png> <image.png> 2013/1/21 Johnson, William Evan <wej@bu.edu<mailto:wej@bu.edu>> Tiger, For question #1: The parametric plots look fine. What you are really looking for here are major distributional deviations from these plots. They actually look pretty good. You'll find only very small differences between the parametric and non-parametric results. Also, no there should not have been a plot for the non-parametric adjustment, as it uses the empirical prior directly, so there are no distributional assumptions to check. Question #2: Check to see if there is a clear "run time" batch effect. If so, compare the two step with the single step. If you have only a few nest run times in each experiment, then the two approaches will be very similar. I would trust the two step approach more than the one- step method. Hope this helps. Evan P.S. I hope you don't mind, I cc'd the bioconductor mailing list where we are now using to answer questions on ComBat. Begin forwarded message: From: hu duan <hu.duan@asu.edu<mailto:hu.duan@asu.edu>> Subject: about ComBat plot for non parametric Date: January 21, 2013 3:52:46 PM EST To: wejlab@gmail.com<mailto:wejlab@gmail.com> Reply-To: hu.duan@asu.edu<mailto:hu.duan@asu.edu> Hi Dr. Johnson I am performing a microarray experiment on cancer research with multiple run times in the two batch of arrays. 1. I have tried your Combat analysis with par.prior=T. The attachment is the plot. There is a difference between red and black line, I am not sure whether it can be called a major violation. I have also tried the nonparametric one. The output was finished but without any plot. Is that normal without a plot?. How should I judge the nonparametric work? 2. Should I first do a Combat analysis in the same batch of slides with multiple run time and then use the adjusted data to do with another batch OR I could do them together? Best Tiger [å† åµŒå›¾ç‰‡ 2] -- Hu Duan (Tiger) Biological Design PhD student Graduate Research Associate Center for Innovation in Medicine The Biodesign Institute, Arizona State University ---------------------------------------------------------------------- ----------------- "MY MIND REBELS AT STAGNATION." -- Sherlock Holmes ---------------------------------------------------------------------- ----------------- -- Hu Duan (Tiger) Biological Design PhD student Graduate Research Associate Center for Innovation in Medicine The Biodesign Institute, Arizona State University ---------------------------------------------------------------------- ----------------- "MY MIND REBELS AT STAGNATION." -- Sherlock Holmes ---------------------------------------------------------------------- ----------------- [[alternative HTML version deleted]]

Hi, I was trying to use cytoscapePlot() from the graphite package, but got the error message below: > library(graphite) > cytoscapePlot(convertIdentifiers(reactome$`Unwinding of DNA`, "symbol")) Loading required package: RCytoscape Loading required package: XMLRPC Error in function (type, msg, asError = TRUE)  : couldn't connect to host In addition: Warning messages: 1: package ‘RCytoscape’ was built under R version 2.15.2 2: package ‘XMLRPC’ was built under R version 2.15.2 What's the problem here? thanks! John [[alternative HTML version deleted]]

Hi Even, Thanks for your response. Best Tiger 2013/1/21 Johnson, William Evan <wej@bu.edu> > See below: > > On Jan 21, 2013, at 4:36 PM, hu duan wrote: > > Hi Evan, > Thanks for your quick response. I may list some interesting result. > > 1. The density plot on the left bottom in previous email shown red > line had a sharp peak while black not. Can you inform me what does that > difference mean? > > Black = density estimate of the batch effects (used directly by > non-parametric adjustments) > Red= Normal/invgamma distributions used for the parametric adjustments > > Ultimately, if the black curve in the first plot is roughly bell shaped, > you won't see much difference in the batch effect adjustments. > > > 1. I just finished analysis of the nonparametric and parametric for a > same data. The following two plots shown here, which the Y axis is the > parametric and X axis is the nonparametric. It seems nonparametric approach > will tend to increase signal. The mean for CV for that is 1.7% while median > is 0.37%. The data looks pretty close. > > > Yes, they are usually very similar unless the black and red lines are > very different from each other. > > > 1. > 2. For the run time, I did a PCA analysis by using all features. The > left pic shown different run times in different color, which clearly shown > run time difference. Right one is after adjust by parametric approach. > Would you consider batch effects have been removed? > > Yes, looks like the batch effects are gone. > > > 1. > 2. If I run Combat analysis in same data and use the adjust data to > run Combat again, do you think the result will be better or it will > overfitting your model? Does the model converge? > > Yes, you might be overfitting a little, but the EB helps reduce the > amount of over-fitting. Also, due to you experimental design, over- fitting > shouldn't be a huge issue. > > > 1. > > Best > Tiger > <image.png> > <image.png> > > > > 2013/1/21 Johnson, William Evan <wej@bu.edu> > >> Tiger, >> >> For question #1: The parametric plots look fine. What you are really >> looking for here are major distributional deviations from these plots. They >> actually look pretty good. You'll find only very small differences between >> the parametric and non-parametric results. Also, no there should not have >> been a plot for the non-parametric adjustment, as it uses the empirical >> prior directly, so there are no distributional assumptions to check. >> >> Question #2: Check to see if there is a clear "run time" batch effect. >> If so, compare the two step with the single step. If you have only a few >> nest run times in each experiment, then the two approaches will be very >> similar. I would trust the two step approach more than the one-step method. >> >> Hope this helps. >> >> Evan >> >> P.S. I hope you don't mind, I cc'd the bioconductor mailing list where >> we are now using to answer questions on ComBat. >> >> >> Begin forwarded message: >> >> *From: *hu duan <hu.duan@asu.edu> >> *Subject: **about ComBat plot for non parametric* >> *Date: *January 21, 2013 3:52:46 PM EST >> *To: *wejlab@gmail.com >> *Reply-To: *hu.duan@asu.edu >> >> Hi Dr. Johnson >> I am performing a microarray experiment on cancer research with multiple >> run times in the two batch of arrays. >> >> 1. I have tried your Combat analysis with par.prior=T. >> >> The attachment is the plot. There is a difference between red and black >> line, I am not sure whether it can be called a major violation. >> I have also tried the nonparametric one. The output was finished but >> without any plot. Is that normal without a plot?. How should I judge the >> nonparametric work? >> >> 2. Should I first do a Combat analysis in the same batch of slides with >> multiple run time and then use the adjusted data to do with another batch >> OR I could do them together? >> Best >> Tiger >> [image: å† åµŒå›¾ç‰‡ 2] >> -- >> Hu Duan (Tiger) >> Biological Design PhD student >> Graduate Research Associate >> Center for Innovation in Medicine >> The Biodesign Institute, Arizona State University >> -------------------------------------------------------------- >> ------------------------- >> "MY MIND REBELS AT STAGNATION." -- Sherlock Holmes >> --------------------------------------------------------------- >> ------------------------ >> >> >> > > > -- > Hu Duan (Tiger) > Biological Design PhD student > Graduate Research Associate > Center for Innovation in Medicine > The Biodesign Institute, Arizona State University > -------------------------------------------------------------- > ------------------------- > "MY MIND REBELS AT STAGNATION." -- Sherlock Holmes > --------------------------------------------------------------- > ------------------------ > > > -- Hu Duan (Tiger) Biological Design PhD student Graduate Research Associate Center for Innovation in Medicine The Biodesign Institute, Arizona State University ---------------------------------------------------------------------- ------ ----------- "MY MIND REBELS AT STAGNATION." -- Sherlock Holmes ---------------------------------------------------------------------- ----- ------------ [[alternative HTML version deleted]]