Plink unable to read plink.dosage.gz file

Question

[paulwjeffries]

I received imputed dosage files from the Michigan Imputation Server (minimac3). The files are vcf format and compressed (gz). I used DosageConvertor to convert the files to plink dosage files; these files are also compressed.

When I try to use a compressed plink dosage file (for example, fileName.plink.dosage.gz) in a linear regression using plink 1.07, plink returns the error "ERROR: Bad format fdr (sic) dosage file, expecting more columns". However, if I uncompress the file and run the same analysis, plink produces no error and completes the analysis. I did include the argument --Zin with the compressed file and omitted the argument when using the uncompressed file.

I used wc to count the number of lines in the compressed file, and the returned line count was what it should be. I counted the number of “words”, and the returned count was correct. The number of columns should equal the number of words / number of lines, since each line should have the same number of words. But since plink is not counting the correct number of columns, does this mean there is a delimiter missing or a delimiter where it should not be.

I believe plink files are white space (space or tab) delimited. Nonetheless, I used sed to change each tab to a single white space and consecutive white spaces to a single white space. But plink is still unable to read the compressed file.

Of course, I can run the analysis with uncompressed files, but it would be nice to keep the files compressed. Can anyone suggest what the problem might be?

All advice is appreciated, Paul

[dnyanada.gokhale]

Hi Paul, I was wondering whether you managed to solve the problem of using compressed dosage files? I am at the same stage right now, having received my dosage files from the Michigan Imputation server. I have used DosageConverter to convert the files to plink dosage and now have a set of compressed plink.dosage files. I need to perform some QC filtering on these in terms of MAF and HWE and was wondering whether it is better to uncompress the files and perform these steps.

Answer 1

[chrchang523]

The problem is that you are still using plink 1.07 for dosage analysis. plink 2.0 can read VCF dosages directly, and supports the full range of linear/logistic regression options on dosage data rather than the limited set offered by plink 1.x --dosage.

[gyang0920]

Hi,

you mean we can use plink run logistic regression for dose.vcf.gz directly? could you share the plink code to run this? i tried many different code. however, no one worked so far...Thank you for your help

[chrchang523]

This depends on how your dosages are encoded in the VCF, but something like

plink2 --vcf [VCF path] dosage=DS --pheno [phenotype file] --pheno-name [phenotype name] --glm

should work.

[gyang0920]

Thank you for your help. So, if i want to add some covariant, I could aff --covar ...? or I need other different command?

[chrchang523]

Yes, use --covar for that.