Unable to liftover my scoring file to GRCh37

[Lasya-Pasumarthy]

Hi, I've been using pgsc_calc for my research, and have been getting low match rates until I realised my scoring file and genomic data are of different genome builds (GRCh38 and GRCh37 respectively). I've been trying to use the --liftover parameter along with the hg38_chain parameter with the chain file being from hg38tohg19.over.chain.gz. But when I run the flow, I get an error: ERROR: Missing --hg19_chain or --hg38_chain with --liftover. I'm not sure what to do, because my scoring file is from PGS Catalog in GRCh38 build and I want to convert it to GRCh37.

[nebfield]

If you're using scoring files from the PGS Catalog that you set with the --pgs_id, --pgp_id, or --trait_efo parameters you don't need to enable liftover. The pgscatalog-download application will automatically fetch scoring files in the correct genome build.

--liftover is for custom scoring files only, that are set with the --scorefiles parameter. This can be helpful if you developed a PGS that isn't in the Catalog and want to test its performance in new target genomes with different genome builds.

When you set --liftover you have to set both --hg19_chain and --hg38_chain parameters, which is what causes your error message.

If you're experiencing low match rates with the correct genome build the most likely problem is that you're using a SNP array that hasn't been imputed or you're using WGS data.

[Lasya-Pasumarthy]

Hi, thank you for clarifying! I'm trying to using the imputed genotypes now, but the available file formats are .bgen, .sample and .bgen.bgi. I'm not sure on what values to put under the format column in the input CSV file.

[nebfield]

The calculator only supports VCFs, plink 1 files (bed / bim / fam), and plink 2 files (pgen / pvar / psam).

It would be best to convert the bgen files to plink 2 format. See here for some discussions about converting bgen files using plink2.

[Lasya-Pasumarthy]

Thanks a lot for providing the links! I went back to using regular genotypes, but I think the number of variants are less. I couldn't complete the pipeline due to an error in match variant and them not meeting the threshold of 0.75. I did get my results now after changing --min_overlap to 0.1 (I remember the website mentioning this might not be the true scores)

[nebfield]

Great 🚀

Your match rates will be higher using imputed data, and are more likely to reproduce the predictive performance reported by score submitters on the PGS Catalog, but I sometimes reduce --min_overlap for debugging and testing.