Error when trying to run pegs_calc on other PGS - Catalogs

[PatrickReck01]

Hey,

I am quite new to PGS Scores and I ran into some issues.

I keep getting the following error:

I get this error when I am trying to use the tool for a vcf file, other than the Test - Sample - Data (the one named 'cineca').
I downloaded the vcf file from 1000Genomes: https://www.internationalgenome.org/data-portal/sample/SC_GMJOL5309922

I also tried it with a dataset I got from Kaggle and transformed into a vcf file, however I got the same error. The Kaggle dataset is this from here: https://www.kaggle.com/datasets/zusmani/mygenome

Interestingly enough, even for the Test Dataset to work, I couldn't use the --pgs_id PGS001229 argument. Instead, I had to use the argument --scorefile '+scoringfilepath+' with scoringfilepath being the path to the scoring file.txt that I found in the folder:
/Users/patrickreck/.nextflow/assets/pgscatalog/pgsc_calc/assets/examples/scorefiles/scorefile.txt

If anyone has any ideas hot to fix this, I would be very thankful. I have been trying to figure this out for a while now, but I don't seem to get any closer.

Best regards
Patrick

[nebfield]

This problem happens when none of the genetic variants in the scoring file can be found in your target (input) genomes. It's typically a problem with the input genotypes or the selected parameters:

• did you accidentally select the wrong genome build?
  • we match variants by looking at genomic coordinates, which change across different genome builds
• do your input genomes have low variant density?
  • we recommend imputing target genomes before using the calculator

If you're looking for some publicly availble data to learn about PGS I suggest the 1000 Genomes dataset, which is helpfully hosted by the plink2 developers. These genotypes have good density and should work with many scores. HAPNEST is also nice, but it's synthetic data and very large.

[PatrickReck01]

Another question that I just thought of:

In the documentation for pgs_calc it says, that the input data should consist of multiple samples, however I am trying to build a web application, where a user can upload his genomic data and receive his pgs scores. That would mean that the input data would only consist of one sample from one individual. Is there any mayor problem with that, or would it work as well?

[nebfield]

Downloading the datasets I mentioned is probably easier than trying to impute the data you have. Genotype imputation can be complicated and confusing.

That's an exciting project! There may be some challenges with low sample sizes. Typically people recommend at least 50 to 100 samples in the target (input) sample. See:

https://www.nature.com/articles/s41596-020-0353-1#Fig2
https://choishingwan.github.io/PRS-Tutorial/target/#sample-size

If you can run pgsc_calc with --run_ancestry low sample sizes are supported because we load missing data from reference populations. However, the ancestry normalisation might be too complicated to include in your project. Good luck!

[PatrickReck01]

I tried to run the datasets (https://www.cog-genomics.org/plink/2.0/resources#phase3_1kg - 1000Genomes phase 3 - GrCh 37) from above, however now I am getting the following error:

Do you have any idea what could cause this? I am not really getting any information out of the error message itself.

As an alternative, I read a forum post (https://www.biostars.org/p/9477182/ ) and it got me thinking, whether I could simply calculate a PGS 'by hand' by calculating the sum over all the SNPs that appear in both the respective pgs-Catalog and the input genome file, weighted by their effect weight. I am assuming, the quality of the PRS score I get this way is lower, however is this option at least somewhat viable?

Alternatively, would it be possible to add my input sample from 1 individual to a bigger dataset with samples from many individuals and use this to calculate the PGS scores? I could then go into results, search the sample I added and read its PGS score.

I also wanted to ask, why there are always datasets with samples from multiple individuals are needed? I read in a paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7612115/ ) its because of quality control mechanisms, however I don't really understand why other individuals genome data would influence the PGS of my single input genome file from 1 individual. I get that large sample sizes is important in GWAS when you are determining which SNPs have an influence on a phenotype, but why is it important when calculating single PGS scores?

I hope I am not bothering you with all my questions. :)
Thanks in advance 👍

[hereagain-Y]

Hi Patrick, when I was trying to use the vcf data from I ran into the same issue as you posted previously, I was wondering did you have any clue to fix it?

[smlmbrt]

@hereagain-Y - what is the exact problem you're having? That no variants match?