Dear Methylomics community,

Using standard ChAMP 2.20.1 package to perform differential methylation analysis, we found very very different results if inputs are A vs B or if they are B vs A, where A and B are sets of properly pre-processed patient M-values. We expected the same (symmetrical) results (those up- in A-vs-B comparison being down- in B-vs-A and so on) but this did not happen. If more experienced colleagues could shed some light on this issue we would greatly appreciate.

MVals_Q <- read.table("../Dados.RVencio/MVals_Q.csv", sep = "\t", header = TRUE)
targets <- read.table("../Dados.RVencio/targets.csv", sep = "\t", header = TRUE)

Values <- dplyr::select(as.data.frame(MVals_Q), matches(c("A|B")))
names(Values) <- gsub ("A_","",names(Values))
names(Values) <- gsub ("B_","",names(Values))
aux1 <- subset(targets, tagets$Sample_Group == "A" | targets$Sample_Group =="B" )
aux1$Classification[aux1$Sample_Group == "A"] <- "A"
aux1$Classification[aux1$Sample_Group == "B"] <- "B"

targets.m <- aux1
myDMP <- champ.DMP(beta = Values, 
                   pheno=targets.m$Classification,
                   adjPVal = 0.01,                       
                   compare.group=NULL,
                   adjust.method = "BH",
                   arraytype = "EPIC")

The screen print is in the following:

A-vs-B

and B-vs-A, the same datset and R-code

Thank you very much in advance for your time

Ricardo Vêncio and Junier Marrero

Hi Ricardo:

I did some test on my computer, with default ChAMP Demo 450K data, but seems the result is the same:

> M[1:5,1:5] # M value 
                  C1         C2         C3        C4         T1
cg00000957  1.935430  2.1515932  2.7772238  2.475677  2.7848624
cg00001349  1.043578  0.7523794  1.0674471  1.290062 -0.1932447
cg00001583 -3.535200 -3.6768762 -0.7424222 -2.399537 -0.3068781
cg00002028 -5.025915 -5.3953388 -4.5841312 -5.154737 -4.7833574
cg00002719 -4.739692 -4.3971369 -3.6020069 -5.423796 -1.7128351
> pheno # phenotype information
[1] "C" "C" "C" "C" "T" "T" "T" "T"
>
> myDMP <- champ.DMP(M, pheno)
...
[ Section 2:  Find Differential Methylated CpGs Start ]

  -----------------------------
  Start to Compare : C, T
  Contrast Matrix
      Contrasts
Levels pT-pC
    pC    -1
    pT     1
  You have found 6023 significant MVPs with a BH adjusted P-value below 0.05.
  Calculate DMP for C and T done.

...

Then I reversely labelled the pheno:

> M[1:5,1:5]
                  C1         C2         C3        C4         T1
cg00000957  1.935430  2.1515932  2.7772238  2.475677  2.7848624
cg00001349  1.043578  0.7523794  1.0674471  1.290062 -0.1932447
cg00001583 -3.535200 -3.6768762 -0.7424222 -2.399537 -0.3068781
cg00002028 -5.025915 -5.3953388 -4.5841312 -5.154737 -4.7833574
cg00002719 -4.739692 -4.3971369 -3.6020069 -5.423796 -1.7128351
> pheno
[1] "T" "T" "T" "T" "C" "C" "C" "C"
>
> myDMP <- champ.DMP(M, pheno)
...
[ Section 2:  Find Differential Methylated CpGs Start ]

  -----------------------------
  Start to Compare : T, C
  Contrast Matrix
      Contrasts
Levels pT-pC
    pC    -1
    pT     1
  You have found 6023 significant MVPs with a BH adjusted P-value below 0.05.
  Calculate DMP for T and C done.

...

I both get 6023 DMPs in both test. My ChAMP version is 2.20.

I think you may check if your pheno is character? Previously I found some factor format pheno may trigger error. Or you may directly send your subset of M value and pheno to champ450K@gmail.com, I will have a look.

Best

Tian