关键词: 主成分分析（PCA）, 随机森林, 高信息量SNP位点, 品种鉴别

Abstract: In order to overcome the difficulties of high dimensional small sample of SNP（Single Nucleotide Polymorphisms） data, while reducing the number of SNP loci for variety identification and meeting the accuracy requirements, this paper proposes a new screening method for SNP loci. Firstly, the main SNP is extracted by PCA, then the random forest with mean accuracy decrease and mean decrease in the Gini index is used to assess the importance of main SNP and train classification model. Finally, the top 48 and 96 sites of importance are selected, these sites as classification feature, classification model is used for variety identification. The model is applied to SNP data from Illumina OvineSNP50 on 6 different sheep breeds. Experiments show that this approach can cut the number of loci from 46013 reduced to 49 or 96, and ultimately the results of classification accuracy can reach more than 97%. The proposed method reduces the number of SNP loci for species identification, and reduces the cost of variety identification.

Key words: Principal Component Analysis（PCA）, random forests, high-informative SNP loci, variety identification