Detecting population structures by independent component analysis

Abstract

In genome-wide association studies (GWAS), detecting population stratification is one of main isuues. As population stratification can produce spurious associations in 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 mainly used for blind source separation, which attempts to distinguish individual signals in situations where multiple signals are mixed. It can treat non-Gaussian data and use higher moments. To determine the population structure, 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, we can detect the population structure and assign individuals to subpopulations. Information about the number of optimal subpopulations can also be obtained. To assess the proposed method, we analyze simulated genotypic data.