Missing value imputation for beta values of Methylation data

Question

[Tanvir Ahamed]

I have download the beta values for methylation data form GEO on different sample. Values form different probes are missing . But instead of delete those probes form my data, i want to impute those values.

impute package and problem

I tried to use impute package though it was developed for microarray data. While using impute package , the computational time is long and facing some problem regarding infinite recursion.

Details about the following problems :

1. Problem :

I got an error while i am running data. Error has explained by the following sample data and code

## Load data 
mdata <- as.matrix(read.table('https://gubox.box.com/shared/static/qh4spcxe2ba5ymzjs0ynh8n8s08af7m0.txt', header = TRUE, check.names = FALSE, sep = '\t')) 

## Install and load library 
source("https://bioconductor.org/biocLite.R") 
biocLite("impute") 
library(impute) 

## sets a limit on the number of nested expressions 
options(expressions = 500000)

## Apply k-nearest neighbors for missing value imputation 
res <-impute.knn(mdata)

Error: protect(): protection stack overflow

2. Problem:

Data : https://gubox.box.com/shared/static/kynad5ajjpqelncdn6djaw7ga35lkvd6.rdata [Note : Big file, 190MB]

library(impute) 
if(exists(".Random.seed")) rm(.Random.seed) 
imputedData <- impute.knn(as.matrix(exp_data))

Error: evaluation nested too deeply: infinite recursion / options(expressions=)?

I will appreciate

1. if anybody can help regarding problem of impute package.
2. Suggest any better way (package/methods) to impute beta values (computationally faster).

Thanks!!

Answer 1

[Tanvir Ahamed]

My Initial solution for "infinite recursion" of impute package :

I have split the input matrix in smaller number by row. I have decreased the number of row in each splited part unitil i dont see the "infinite recursion" error message. After that , i have merged all the imputed matrix of splited parts.

R code :

myimpute <- function(data,clus = 2000) # clus = Number or row in splited matrix
{
library(impute)
data <- data.frame(data)
row_data <-nrow(data)
ind<-as.factor(c(gl(round(row_data/clus)-1,clus),rep(round(row_data/clus)-1+1, nrow(mm)-length(gl(round(row_data/clus)-1,clus)))))
newMat <- split(data, ind)
res <- lapply(newMat,function(x)impute.knn(as.matrix(x)))
res <- lapply(res,"[[","data")
res <- do.call(rbind, res)
res
}

Any one has any thought on this , please share .