"WARNING::*number*% GWAS SNPs (z_snp$id) are in weights!"

[ameyers99]

Hi there,
Thanks for a great package.
I have been trying to test cTWAS on complex genomic regions and get the warning message "WARNING::number% GWAS SNPs (z_snp$id) are in weights!" when many variants in a given region are non-zero in the corresponding expression prediction models. Before exploring further, I was hoping you may be able to explain why this warning message occurs?
Thank you,
Aaron

[kevinlkx]

Thanks for your interest in cTWAS. About the warning message, it suggests some potential issue with the input data. We typically expect that only a small fraction of GWAS SNPs are used in the weights of the prediction models. We give this warning when more than half of the GWAS SNPs are in weights. From your warning message, what is the percentage of your GWAS SNPs are in weights?

[ameyers99]

Thank you for the response. In my case, the warning message indicates that around 80-90% of variants are included in the weights. If, hypothetically, I were studying a complex genomic region that genuinely contains few variants, and most (or even all) of those variants are used in prediction models, would cTWAS in principle still be able to reliably separate gene and variant effects? Or does such high overlap inherently limit the models ability to distinguish these signals?
Thank you.

[kevinlkx]

Thanks for your question. This would not necessarily lead to problems. Even if many variants are included in the weights, the vast majority of variants often have small weights, so their effects are generally small. In other words, the “effective” number of variants in the weights are often relatively small compared to the total number of variants. Having said that, it would be helpful to filter variants with small weights in the prediction models before running cTWAS.

We had some users whom incorrectly removed GWAS variants that are not in weights before running cTWAS. So we gave this warning for users to be careful about potential problems with the input data.

[ameyers99]

Hi Kevin,
Thank you for your response and for explaining the warning message – very helpful!
Best,
Aaron