Highly inflated p-value in Tractor GWAS

[ryshi06]

Hi,

I tried to run Tractor on 1765 subjects. I followed all the steps, and two population group AFR and EUR. However, I found high inflation in the output GWAS, with the genomic inflation factor 0.5, also the Q-Q plot looks weird. Can you please give me some hint on this? I appreciate your help!

Best, Ruyu

[nirav572]

Hi Ruyu,

Thank you for reaching out! A couple of things to clarify/check:

• Has your data already been QC’d, retaining only high-quality variants?
• Is your local ancestry inference working correctly and using the appropriate reference panels?

One way to validate this is by generating ADMIXTURE global proportions and comparing them to your results — you should observe high concordance.

If everything looks good in these preliminary steps, are you controlling for the right factors using covariates. You could run a standard GWAS with your phenotype of interest and see if you find any significant hits in the process? It’s also possible that your sample size may currently be too small to identify any associations.

Let us know if we can help in any other way. Nirav

[ryshi06]

Hello Nirav,

Thank you for your response! Could you please provide more details regarding the validation option?

In the meantime, I wanted to share that out of the 1765 samples, only about 290 are of African ancestry, while the majority are of European descent. The Q-Q plot I sent earlier was from the AFR ancestry GWAS. In contrast, the Q-Q plot for the EUR population (MAF >1%) looks good, and the top hits align with my standard GWAS results. Based on this, is it reasonable to conclude that the small sample size for the AFR population might limit the power, even with Tractor, to detect significant associations? Additionally, I noticed that Tractor does not seem to perform well with rare variants (MAF <1%), which I find puzzling. Could you please clarify why this might be the case?

Best, Ruyu

[JasonTan-code]

Hi, lambda = 0.5 means it's deflated, not inflated, which means you might have too many type 2 errors (false negatives) for AFR. Indeed Tractor is not optimized for imbalanced ancestry proportions, and variants with low AF. The results might just be a consequence of small sample size

[nirav572]

Looks like the issue has been resolved, so I’ll close this for now. Reopen if needed.