Closed mkierczak closed 2 years ago
Hi @mkierczak, It is known that the beta estimation based on the score test statistics is biased, especially for rare variants. We now just have released a new version 1.0.0 of SAIGE/SAIGE-GENE, in which we implemented the the effect size estimation through the Firth's Bias-Reduced Logistic Regression. Please feel free to try it and open the issue in the new github. https://github.com/saigegit/SAIGE Thanks, Wei
Hi! I am working with simulated data where I know the exact effect size. When running SAIGE SPA tests, I see my simulated signals (single common variant)correctly being the ones with the lowest p-value but the BETAs are way off from what I am simulating. I tried to change kernel to
linear
as well as setting Beta distribution params for common and rare alleles toBeta(1,1)
with no improvement. In my toy example, effects are supposed to be around 1. BETAs I observe, are really spread with no systematic error as far as I can see. Could you possibly have some hints for me?