Exploratory analysis for detecting population structures by iterative pruning based on independent component analysis

Abstract

As population stratification can produce spurious associations in genome-wide association studies (GWAS), it is necessary to find hidden structures and assign individuals to subpopulations in advance. We suggest an exploratory approach for population structure analysis based on independent component analysis (ICA). ICA is unsupervised approach to identifying and separating mixed sources from observed signals with little prior information. We first reduce the dimensionality of samples by projecting the data to a lower-dimensional subspace built by ICA. The samples are then bisected using fuzzy clustering. Repeating this procedure until some predetermined stopping criterion such as negative entropy, we can detect the population structure and assign individuals to subpopulations. Information about the number of optimal subpopulations can also be obtained. We analyse simulated genotypic data with different degrees of structure. We compare the proposed method to other methods. Real data from the HapMap project are also analysed for illustration.