Remove NaNs from ExpressionSet class object

Question

[Morteza Hadizadeh]

Hi everyone,

Sometimes in microarray analysis, some datasets contain NaN values, I need to remove NaNs from gset to ex matrix will be without NaNs too.

How to remove NaNs from ExpressionSet (gset)?

I would be very pleased if you could guide me.

Answer 1

[zx8754]

We need to use Biobase::exprs() to access matrix data, then assign as usual, see this reproducible example:

Reproducible example from the manual - An Introduction to Bioconductor’s ExpressionSet Class

library(Biobase)

dataDirectory <- system.file("extdata", package="Biobase")
exprsFile <- file.path(dataDirectory, "exprsData.txt")
exprs <- as.matrix(read.table(exprsFile, header=TRUE, sep="\t",
                              row.names=1,
                              as.is=TRUE))

minimalSet <- ExpressionSet(assayData=exprs)

# check the data
exprs(minimalSet)[1:3, 1:3]
#                        A         B        C
# AFFX-MurIL2_at  192.7420  85.75330 176.7570
# AFFX-MurIL10_at  97.1370 126.19600  77.9216
# AFFX-MurIL4_at   45.8192   8.83135  33.0632

# introduce some NaN
exprs(minimalSet)[1:2, 1:2] <- NaN

# check the data
exprs(minimalSet)[1:3, 1:3]
#                       A       B        C
# AFFX-MurIL2_at      NaN     NaN 176.7570
# AFFX-MurIL10_at     NaN     NaN  77.9216
# AFFX-MurIL4_at  45.8192 8.83135  33.0632

# replace NaN with zero
exprs(minimalSet)[ is.nan(exprs(minimalSet)) ] <- 0

# check the data
exprs(minimalSet)[1:3, 1:3]
#                       A       B        C
# AFFX-MurIL2_at   0.0000 0.00000 176.7570
# AFFX-MurIL10_at  0.0000 0.00000  77.9216
# AFFX-MurIL4_at   45.8192   8.83135  33.0632

Answer 2

[Friederike]

You should be able to subset the ExpressionSet like a base-R matrix, i.e.

ExpressionSet[is.na(ExpressionSet)] <- 0

If that doesn't work, try the same approach with exprs(ExpressionSet) or assayData(ExpressionSet). See this website for more ways of replacing NA's.

[Morteza Hadizadeh]

Hi Friederike,

Thank you for your guidance and useful link that you introduced. I mean NaNs, not NAs, NaNs is different from NAs.

[Ram]

Change is.na in Friederike's command to is.nan and you'll replace NaNs instead of NAs

[Morteza Hadizadeh]

Hi,

Thank you for your attention, while I use is.nan instead is.na, encounter the following error:

> ExpressionSet[is.nan(ExpressionSet)] <- 0
Error in is.nan(ExpressionSet) : 
  default method not implemented for type 'closure'

[archana.bioinfo87]

Hi,

try this code maybe it will work for you

ExpressionSet[ExpressionSet == NaN] <- 0

[Ram]

Hi,

Comparing to NaN with == won't work unless you've redefined ==.

> NaN == NaN
[1] NA
> is.nan(NaN)
[1] TRUE

[Morteza Hadizadeh]

Hi, Thank you for your answer. I use the above code and encounter the following error:

> ExpressionSet[ExpressionSet == NaN] <- 0
Error in ExpressionSet == NaN : 
  comparison (1) is possible only for atomic and list types

[Ram]

I think we have a bit of misunderstanding here. Is your object called gset or ExpressionSet? If it's gset, you should be using gset[is.nan(gset)]<-0.

If that doesn't work, Friederike's exact code could work for you. is.na() matches NA as well as NaN.

If gset contains NaN but not NA, you can simply do gset[is.na(gset)] <- 0. If your object contains both NAs and NaN, (and gset is a data.frame like object), you can do something like:

gset.tmp <- gset #So we don't end up corrupting the original object
gset.tmp[do.call(cbind, lapply(gset.tmp, is.nan))] <- 0 # This should replace NaNs with 0s

Please try both approaches above and let us know how that works out for you.

[Morteza Hadizadeh]

Hi, Thanks for your follow up. gset class is ExpressionSet.

> class(gset)
[1] "ExpressionSet"
attr(,"package")
[1] "Biobase"

The above-mentioned codes are for data.farame object, not ExpressionSet. See an example of the above codes below.

> gset[is.nan(gset)]<-0
Error in is.nan(gset) : default method not implemented for type 'S4'
> gset[is.na(gset)] <- 0
Error in gset[is.na(gset)] <- 0 : object of type 'S4' is not subsettable
In addition: Warning message:
In is.na(gset) : is.na() applied to non-(list or vector) of type 'S4'
> gset.tmp <- gset
> gset.tmp[do.call(cbind, lapply(gset.tmp, is.nan))] <- 0
Error in as.list.default(X) : 
  no method for coercing this S4 class to a vector
> ExpressionSet[ExpressionSet == NaN] <- 0
Error in ExpressionSet == NaN : 
  comparison (1) is possible only for atomic and list types

[Ram]

This needs for someone to read the manual and figure out what sort of class ExpressionSet is.

From a quick read, it looks like the numeric data in an ExpressionSet object is stored in its AssayData slot. You should be able to use:

dfES<-as(gset, 'data.frame');
dfES[do.call(cbind, lapply(dfES, is.nan))] <- 0

or

aD <- assayData(gset);
aD[do.call(cbind, lapply(aD, is.nan))] <- 0

P.S.: Please do not use ExpressionSet like you would a variable, it's not. It's the class of a variable. Using it as a variable is like using data.frame[is.na(data.frame)]<-0, it shouldn't work, and if it does, there's a bigger problem with the code.

Answer 3

[Morteza Hadizadeh]

Thanks for all the suggestions and valuable guides. I convert ExpressionSet into a Matrix, following replace NaNs with zero and convert the matrix into ExpressionSet again.

> ex <- exprs(gset)
> ex[is.nan(ex)] <- 0
> exprs(gset) <- ex
> dim(ex)
[1] 47323    86

> dim(gset)
Features  Samples 
   47323       86

[Friederike]

Just to be precise, some advice on your wording: The ExpressionSet object (in your case named "gset") contains a matrix of values, which can be accessed via exprs() (as you show).

[Morteza Hadizadeh]

Thank you and everyone else, for sharing your wisdom with me.

[Kevin Blighe]

Thank you for posting the answer. Can you please up-vote and / or Accept the other answers that have assisted you?

[Morteza Hadizadeh]

Thanks for your follow up. The answers were valuable and helpful.

[Ram]

Then you should accept the answers that solved your question and/or pointed you in the right direction.