Inversion of k-Nearest Neighbours Algorithm for Extracting SNPs Discriminating Human Populations.

With the development of new technologies, many multi-class and high dimension data have been accumulated in the biology field . The data contains much useful information. But how to mine the information is a hard problem. The international project (HapMap) has collected much SNP (Single-nucleotide polymorphism) data of individuals for different human races, however, which SNPs lead to the differences between human races is unknown. If these SNPs are extracted, it will be very useful for genetic studies. In the paper, a novel algorithm is proposed to extract the SNPs discriminating human races. The algorithm adopts an inversion of k-nearest neighbours algorithm (IKNN) which uses an iterative procedure to modify the weights of each SNP to make every individual belong to the same population as its k-nearest neighbours. When the weights convergences, most weights of the SNP site are zero which means that these SNPs are noises for classification. The rest SNPs are important for classification. We validate our method on HapMap data, IKNN has a better performance than neural network algorithm and KNN algorithm.