When running the sva function, I get an error "Error in density.default(x, adjust = adj) : 'x' contains missing values".

I have a data frame with raw RNA-seq counts with sample names in columns and gene names in rows, and another data frame describing the condition of each sample. There is only one condition column and it has 8 possible states (I cannot post the actual dataset). Here is my code:

library(sva)

mycoldata <- read.delim("contrast.txt", header = TRUE, sep = "\t")

mycountdata <- read.delim("expr_table.tsv", header = TRUE, sep = "\t")
mycountdata$checksum <- NULL
mycountdata$score <- NULL
df <- data.matrix(mycountdata)

# build the FULL model matrix
mod = model.matrix(~as.factor(condition), data=mycoldata)

# build the NULL model matrix
mod0 = model.matrix(~1,data=mycoldata)

n.sv = num.sv(df,mod,method="leek")
n.sv

# estimate the surrogate variables
svobj = sva(df,mod,mod0,n.sv=n.sv)

n.sv for this data set is 14. When I run the last command, I get the following error message:

> svobj = sva(df,mod,mod0,n.sv=n.sv)
Number of significant surrogate variables is:  14 
Iteration (out of 5 ):Error in density.default(x, adjust = adj) : 'x' contains missing values
In addition: Warning message:
In pf(fstats, df1 = (df1 - df0), df2 = (n - df1)) : NaNs produced

I found some posts describing what looks to be the same error [1][2][3], however none of these seem to have led to an accepted explanation of this error or a way to resolve this (approaches described were: just reducing the number of n.sv until it works, removing genes with low counts until it works).

[1] https://www.biostars.org/p/198820/

[2] https://support.bioconductor.org/p/78142/

[3] https://stackoverflow.com/questions/43101585/error-when-generating-the-sva-object-using-package-sva-in-r

I had a similar problem. I found that my factor variables read in as characters (which doesn't matter for some packages but it does for this one). Once I changed all my variables labeled as characters to factors the function worked.

I found the reason, because some gene expression levels are 0 in all samples. you can add these code to check the data:

exp_mean <- apply(exp,1,mean)

filter_gene <- rownames(subset(exp_mean ,exp_mean ==0))

exp_mean2 <-exp_mean [-which(rownames(exp_mean ) %in% filter_gene),]

You can try to remove these genes.

This same error is returned also when the number of requested SVs is too high, when manually specifying with n.sv=....

I had the same problem. When I did not setup 'n.sv' parameter, like svobj = sva(df,mod,mod0), the code worked and n.sv was calculated when running this code.

However, I did not whether my running was right.

Hope to be replied for cross-checking.