atac_dim_reduction() may throw a BLAS error: Fix by removing NAs

[hypercompetent]

Hi Andrew and Darren,

I was testing the atac_dim_reduction() function today on a subset of my data, and got a mysterious BLAS error at the SVDtsne = irlba(tf_idf_counts, 50, 50, maxit=1000) step:

Error in irlba(tf_idf_counts, 50, 50, maxit = 1000) : 
  BLAS/LAPACK routine 'DLASCL' gave error code -4

After a little troubleshooting, I found that I had some NA values in my tf_idf_counts matrix. I added this line, and it runs well:

tf_idf_counts = nfreqs * log(1 + ncol(ncounts) / Matrix::rowSums(ncounts))

tf_idf_counts@x[is.na(tf_idf_counts@x)] <- 0 ## New line to remove NAs

## Do SVD
set.seed(0)
SVDtsne = irlba(tf_idf_counts, 50, 50, maxit=1000)

I'm not sure if this will come up for every dataset - a larger set of cells it didn't seem to throw this error - but I thought I'd post here in case anybody else hits the same issue.

Thanks for sharing these functions!
-Lucas