Table 1. Performances of feature selection methods.

Associations of the input dataset without or with population stratification correction by genomic control and Eigenstrat (with 5 and 10 principal components considered) compared with SNPClust.

Feature selection method	Number of p-values < 5%	Number of FDR q-values < 5%	Number of FDR q-values in HLA < 5%
(none)	97,066	39,555	951
Genomic control	18,267	271	232
Eigenstrat 5	22,605	117	64
Eigenstrat 10	15,269	28	9
SNPClust	222	156	38