Comparison of three methods for obtaining principal components from familydata in genetic analysis of complex disease

Three multivariate techniques used to derive principal components (PCs) fro m family data were compared for their ability to model family data and powe r to detect linkage. Using the simulated data from Genetic Analysis Worksho p 12, the five quantitative traits were first adjusted for age, sex, and en vironmental factors 1 and 2. Then, standard PCs, PCs obtained from between- family covariance, and PCs obtained from within-family genetic covariance w ere derived and subjected to multivariate sib pair linkage analysis. The st andard PCs obtained from the overall correlation matrix allowed identificat ion of key features of the true genetic model more readily than did the oth er methods. For detection of linkage, standard PCs and PCs obtained from th e between-family genetic covariance performed similarly in terms of both po wer and type I error, and both methods performed better than the PCs obtain ed from within-family genetic covariance. (C) 2001 Wiley-Liss, Inc.