Question: Limma: Warning Messages -- rlm -- Is there a way to fix it?
0
2.8 years ago by
alakatos110
United States
alakatos110 wrote:

Hello All,

At the end of my DE  analysis  in limma I received the following warning message: "There were 50 or more warnings (use warnings() to see the first 50)"

warnings()Warning messages:
1: In rlm.default(x = X, y = y, weights = w, ...) : some of ... do not match  'rlm' failed to converge in 20 steps etc.

I tried  to go around  the problem by increasing the maxit option to 50. Unfortunately, it did not solve the problem.

Would you please advise if there is a solution to this problem or  shall I neglect it ?

Anita

My code:

Elistraw = read.idat(idatfiles, bgxfile)

y <- neqc(Elistraw)​

des <- model.matrix(~ dx + batch + sex + age + source + pluritestscore + depression)
des

(Intercept) dxDem dxMCI batch2 batch3 batch4 sexFemale age sourcePBMC pluritestscore depressionDep
1           1     0     1      0      0      0         1  80          0         21.261             0
2           1     1     0      0      0      0         0  81          0         19.290             0
3           1     0     0      0      0      0         0  83          0         23.087             0
4           1     0     0      0      0      0         1  67          0         21.847             0
5           1     0     1      0      0      0         1  67          0         23.738             0
6           1     1     0      0      0      0         0  75          0         23.901             1

fit <- eBayes(lmFit(y,des,trend = TRUE, method="robust"))

There were 50 or more warnings (use warnings() to see the first 50)

fit <- eBayes(lmFit(y,des,trend = TRUE, method="robust", maxit=50))​

There were 50 or more warnings (use warnings() to see the first 50)

> sessionInfo()
R version 3.3.1 (2016-06-21)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252    LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                           LC_TIME=English_United States.1252

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base
other attached packages:
[1] Hmisc_4.0-0           foreach_1.4.3         reshape_0.8.6         ggplot2_2.2.0         Formula_1.2-1
[6] survival_2.40-1       lattice_0.20-34       flashClust_1.01-2     dynamicTreeCut_1.63-1 cluster_2.0.5
[11] dplyr_0.5.0           illuminaio_0.12.0     magrittr_1.5          limma_3.30.6
loaded via a namespace (and not attached):
[1] Rcpp_0.12.8         RColorBrewer_1.1-2  plyr_1.8.4          iterators_1.0.8     tools_3.3.1         digest_0.6.10
[7] rpart_4.1-10        base64_2.0          tibble_1.2          gtable_0.2.0        htmlTable_1.7       Matrix_1.2-7.1
[13] DBI_0.5-1           gridExtra_2.2.1     stringr_1.1.0       knitr_1.15.1        grid_3.3.1          nnet_7.3-12
[19] data.table_1.10.0   R6_2.2.0            foreign_0.8-67      latticeExtra_0.6-28 MASS_7.3-45         codetools_0.2-15
[25] htmltools_0.3.5     scales_0.4.1        splines_3.3.1       rsconnect_0.6       assertthat_0.1      colorspace_1.3-1
[31] stringi_1.1.2       acepack_1.4.1       lazyeval_0.2.0      openssl_0.9.5       munsell_0.4.3      

limma warning message • 1.3k views
modified 2.8 years ago by Gordon Smyth38k • written 2.8 years ago by alakatos110
Answer: Limma: Warning Messages -- rlm -- Is there a way to fix it?
1
2.8 years ago by
Gordon Smyth38k
Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia
Gordon Smyth38k wrote:

You may be able to just ignore this warning. There is no easy fix anyway. It might be affecting only a subset genes, although it is admittedly hard to check this from the error message.

I was the one who added the robust option to lmFit() more than 12 years ago, but over the years I have found it to be very seldom necessary. In particular, I've never found it be necessary for Illumina microarray data. So I would suggest you simply try the analysis without using this option. The way that limma analyses the data on the log-intensity scale is quite robust anyway. limma has other ways to deal with outlier samples or outlier genes.