How to deal "missing value error" on imputed genotypes data

[panqinglzmc]

I am using the demonstration data "penncath" from the tutorial of the R package [bigsnpr]. I have verified that there are no missing values in the imputed obj$geno_imputed data, both by SNP and samples. However, when I use the big_SVD function for principal component analysis, I still encounter an error: "You can't have missing values in 'X'", as shown below. I would be very grateful if someone could help me identify where I might have made a mistake.

[privefl]

• using the demonstration data "penncath" from the tutorial of the R package [bigsnpr]

  Not sure which data you're talking about??

• Could you show snp_stats[, 1:5] please? (including the rownames)

• Not related to this issue, but I would recommend using snp_autoSVD() instead of big_SVD() for genotype data.

[panqinglzmc]

Thanks Florian for the quick turnaround! And I must apologize for not explaining clearly. I am using the imputed genotype data from the GWAS tutorial: Imputation available at this link. The data was generated using the following code:

The snp_stats[, 1:5] output is as follows, identical to what is shown in the tutorial.

However, when passing this obj$geno_imputed data to the big_SVD function in the subsequent code in the GWAS tutorial: Population structure (available here), it results in the error mentioned above.

### But I am very happy to say that following your suggestion to use snp_autoSVD() instead of big_SVD() for genotype data, the problem has been solved, and the code now runs smoothly, as shown below.

Thank you so much, Florian. Your suggestion has been incredibly helpful.

[privefl]

This is good that you found some workaround.
But the initial issue is not really fixed.
I have some idea what's going on.
Could you please confirm your packageVersion("bigstatsr")?

[panqinglzmc]

My ‘bigstatsr’ version is 1.5.12.

Ps. My 'bigsnpr' version is 1.12.2. I loaded the bigsnpr package, and the bigstatsr package was automatically loaded along with it. Subsequently, I used the following two functions: snp_fastImputeSimple {bigsnpr} and snp_autoSVD {bigsnpr}.

[privefl]

What do you get if you run this reproducible code?

zip <- runonce::download_file(
  "https://d1ypx1ckp5bo16.cloudfront.net/penncath/penncath.zip",
  dir = "tmp-data")
unzip(zip, exdir = "tmp-data", overwrite = FALSE)

library(bigsnpr)
snp_readBed("tmp-data/data/penncath.bed")
penncath <- snp_attach("tmp-data/data/penncath.rds")
penncath$geno_imputed <- snp_fastImputeSimple(Gna = penncath$genotypes,
                                              method = "mode",
                                              ncores = nb_cores())

big_SVD(penncath$geno_imputed, big_scale(), k = 10)

For me, it runs forever because there are some variables with no variation that prevent convergence (which now errors with v1.5.14).
But I don't get the error about missing values (with both v1.5.12 and v1.5.15).

[panqinglzmc]

I ran the code, and come out the same error.

[privefl]

I cannot reproduce the issue, and I have no idea what's going on :/
Is this the only function where you have this issue? (e.g. if you also try running big_univLinReg(penncath$geno_imputed, rnorm(1401), ind.col = 1:100))

PS: You should try not to change the working directory; use RStudio projects and stick with the working directory of the project.