NetRep error: matrices cannot have non-finite or missing values

[abadgerw]

@sritchie73 thank you for a great tool! I am trying to run NetRep on my data. It has missing values but has been filtered so each gene is in >50% of samples so that I can run WGCNA. I have been using pairwise complete obs for tabulating correlations and adjacency matrices. However, when running ModulePreservation I am getting the following error:

NetRep Error: matrices cannot have non-finite or missing values

I thought based on the fact that "caution should be used when running NetRep on sparse data" that it wouldn't error if missing values were present.

I am bootstrapping my reference data to generate the test sets and wasn't sure if that could be contributing to the issue.

Any insight you can provide would be very helpful.

[sritchie73]

From memory, I think NetRep should work with missing values in the gene expression matrix, but expects the correlation and adjacency matrices to have all non-missing values. In the context of sparse networks, edges that are not present would all have 0 values rather than NAs.

In the context of filtering genes to these in >50% of samples, I think the main case you have to think about is how you want to handle cases where the gene is present in >50% of the samples in one dataset but not the other. Your two options are (1) treating the correlations as 0 for that gene in the dataset where the gene is present in < 50% of the samples, or (2) filtering your datasets to include only the subset of genes that are present in both samples.

The approach you go with really depends on what "Gene is not present in >50% of samples" actually means. If this means the gene is probably not expressed in that dataset, I'd go with option 1, but if it means that the measurement of the gene is unreliable in that dataset I'd go with option 2.

[abadgerw]

Thanks, @sritchie73! I've confirmed I have no missingness in the adjacency or correlation matrices. I only have missing values in the expression matrix but still can't get the function to work. Any insights into why this may be?