Batch Effect normalization with Combat generates error
I have 6 batches (10 samples in each) from 6 arrays that we want to normalize for batch effect with Combat. When I run the script below, I encounter an error, any idea how to fix this?
I'm editing the current code with all modification and current error message that I get.
library(sva)
dat = read.csv("Combat_matrix_input.csv");
sif = read.csv("sif.csv");
batch = sif$Batch
head(batch)
[1] Batch1 Batch1 Batch1 Batch1 Batch1 Batch1
Levels: Batch1 Batch2 Batch3 Batch4 Batch5 Batch6
modcombat = model.matrix(~1, data=dat)
dat <- as.matrix(dat)
combat_edata = ComBat(dat=dat, batch=batch, mod=modcombat, par.prior=TRUE, prior.plots=FALSE)
Found 6 batches
Error in cbind(batchmod, mod) :
number of rows of matrices must match (see arg 2)
# Here I'm trying with batch as.vector
batch = as.vector(sif$Batch)
combat_edata = ComBat(dat=dat, batch=batch, mod=modcombat, par.prior=TRUE, prior.plots=FALSE)
Found 6 batches
Error in cbind(batchmod, mod) :
number of rows of matrices must match (see arg 2)
Here I display the output of column Batch in sif
# I tried both batch as.vector as you guys recommended
batch = as.vector(sif$Batch)
batch = sif$Batch
# Both batch as.vector or regular variable return the error in combat before, and return the same head()
head(sif$Batch)
[1] Batch1 Batch1 Batch1 Batch1 Batch1 Batch1
Levels: Batch1 Batch2 Batch3 Batch4 Batch5 Batch6
1 answer
I'm pretty sure that your variable should be numeric to indicate batches. So change sif$Batch to numeric
• 0 views
•
link
updated
by
Ram
•
written
by
andrew.j.skelton73
•
written
by
andrew.j.skelton73
ok, change the 'mod' parameter in your combat call to 'NULL' - So,
combat_edata = ComBat(dat=dat, batch=batch, mod=modcombat, par.prior=TRUE, prior.plots=FALSE)
is changed to
combat_edata = ComBat(dat=dat, batch=batch, mod=NULL, par.prior=TRUE, prior.plots=FALSE)
• 0 views
•
link
updated
by
Ram
•
written
by
andrew.j.skelton73
•
written
by
andrew.j.skelton73
Log in to answer this question.
More posts like this
-
How to handle Single-Level Batch Variable for Batch Correction in RNA-Seq Data
written by Nelo •
-
Is it necessary to standardize Combat in R?
written by tkmt •
-
Adjusting for batch effect and covariates with ComBat
written by juara •
-
SVA combat error covariate is confounded with batch
written by CL •
-
In SVA, the percent of biological variable of interest didnot change after adding SV1.
written by dhwani2410 •
-
Help with multiple batch effects
written by fp89 •
-
svaseq and combat
written by Folder40g
-
R: Error when using ComBat
written by kkj3083 •
-
what is the meaning of covariates in ComBat
written by LJ
-
How to use Combat in R??
written by madkitty
Can you show us the output of
@andrew.j.skelton73: I updated the original post with the output of
head(sif$Batch). Thanks :)Will
as.vector(sif$Batch)help?Just found my script. This was what I did with ComBat
I was trying to merge RNA Seq data with microarray.
mydatais my microarray data andvsdis my RNA-Seq data whereas data.input is the merge data frameThanks Sam, though in our case we only have genes in rows and samples in columns. The sample info file contains only array, sample_name and batch. Without any condition, how should I write mod? I don't understand what should go under
~as.factor()@Sam: I tried that
Unfortunately the result is the same..
According to page 6 of the user manual,
The deviation of this code and yours is that instead of
data=pheno, you usedata=dat. If you look at my example script, you should note that thephenois the phenotype matrix. What I suspect is the data in the mod should contain the phenotype and the batch instead of the count table. try and see ifsolves the problem? That's just my guess though
Hi Sam, Thanks again for your answer. When replacing for data=sif in modcombat, I get another type of error (see below). The manual isn't useful at all .. I'm really desperate to find a solution :(
Can you show us how your
siflooks like?Actually, I gave it another try, and indeed your suggestions worked well. Just for those wondering, the manual isn't clear, MOD should contain SAMPLE INFO FILE and COVARIATE information as a data.frame. Thanks Sam!
Good to know that it works