GWAS Bench: Manhattan, QQ, lambda, power

Study

Seed. The same one rebuilds the same study.

Cases

Controls

Markers tested

About a million common variants are independent in the human genome, which is where 5e-8 comes from.

Significance line

What is really there

True associations

Set to 0 for a study with no true associations. The slider stops at 50; type a larger number for a polygenic trait.

Odds ratio

Per allele. Most common-variant hits are under 1.2.

Causal MAF

Population structure

Split the sample into two populations

Ancestry Fst

0.001 is two European countries, 0.01 Europe against South Asia, 0.05 and up is continental. Logarithmic slider.

Case imbalance

Share of cases drawn from population 1. Controls get the mirror image, so 0.5 means no confounding.

Correction

Two populations differ in allele frequency by Fst, and one of them supplies most of the cases. Any marker that differs between them now differs between cases and controls, for no biological reason.

Genomic control divides every chi-square by lambda. It removes the inflation and shrinks the real hits by the same factor.

Adjust for ancestry is the idealised version of the principal-component covariates a real study uses: here the tool knows each person's population, which a real study has to estimate.

Replication tab

Replication N

Total, split evenly into cases and controls.

Presets

Manhattan plot

mark the true hits