Interpretation of the axes of genetic variation is crucial before testing their association with disease. In a homogeneous population, the first components of genetic variation may reflect local extended LD patterns rather than genome-wide structure, such as that reported in the main WTCCC experiment. Clearly, if the region of extended LD contains causal variants, regressing out the corresponding axis of genetic variation will prevent their detection. This is likely to be true for autoimmune diseases in the MHC, for example, an extended region of strong LD harboring many established associations. Apparent axes of genetic variation may also reflect “batch effects,” resulting from the use of different genotyping platforms or calling algorithms between cohorts, which will be important to take account of in the same way as population structure, in order to reduce the inflation in false-positive error rates.
