Hi Niles Thanks for reporting this. The warning message was produced when the "smoothQuantileNormalization" tries to use "rlm"(in MASS package) to detect outliers in the low or high intensity range. Sometimes, "rlm" fails to converge in these regions, especially for low intensity region. Since the number of low intensity probes (usually these are failed probes) are very low for DNA methylation data, it usually should not affect the overall processing. Probably, I should suppress this kind of warning messages. Pan On Wed, Jun 13, 2012 at 9:19 AM, Niles Oien <nilesoiencu at="" gmail.com=""> wrote: > > I am a research assistant at the University of Colorado. > > We have some methylation data that we are trying to normalize by calling > lumiMethyN() in the lumi package, however, we are finding that we get the > message that "rlm did not converge in 20 steps". We have tried both quantile > and ssn methods, but still get this message. I do notice that instead of > specifying "quantile" or "ssn" as the method, one can specify a user defined > function?whose input and output should be a intensity matrix (pool of > methylated and unmethylated probe intensities). Do you think that is what we > should be doing? And could you elaborate about how we do that? > > Thanks, any ideas you have would be appreciated - > > Niles Oien. >