Rank and order: evaluating the performance of SNPs for individual assignment in a non-model organism

• Published in: PloS one 2012-11, Vol.7 (11), p.e49018-e49018
• Main Authors: Storer, Caroline G, Pascal, Carita E, Roberts, Steven B, Templin, William D, Seeb, Lisa W, Seeb, James E
• Format: Article
• Language: English
• Subjects: Alaska; Analysis; Animals; Biological evolution; Biology; Chromosomes; Computer simulation; Data processing; Ecological monitoring; Ecology; Endangered & extinct species; Evolution; Fish; Fishery sciences; Fishes; Fishing; Genetic Loci - genetics; Genetics; Genomes; Geography; Heterozygote; Innovations; Laboratories; Linux; Loci; Models, Animal; New technology; Oncorhynchus keta; Oncorhynchus nerka; Organisms; Panels; Polymorphism; Polymorphism, Single Nucleotide - genetics; Population; Principal Component Analysis; Probability; Ranking; Rankings; Reproducibility of Results; Salmon; Salmon - genetics; Sample Size; Securities analysis; Single nucleotide polymorphisms; Single-nucleotide polymorphism; Specimen Handling; Statistics, Nonparametric; Studies
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0049018

Single nucleotide polymorphisms (SNPs) are valuable tools for ecological and evolutionary studies. In non-model species, the use of SNPs has been limited by the number of markers available. However, new technologies and decreasing technology costs have facilitated the discovery of a constantly increasing number of SNPs. With hundreds or thousands of SNPs potentially available, there is interest in comparing and developing methods for evaluating SNPs to create panels of high-throughput assays that are customized for performance, research questions, and resources. Here we use five different methods to rank 43 new SNPs and 71 previously published SNPs for sockeye salmon: F(ST), informativeness (I(n)), average contribution to principal components (LC), and the locus-ranking programs BELS and WHICHLOCI. We then tested the performance of these different ranking methods by creating 48- and 96-SNP panels of the top-ranked loci for each method and used empirical and simulated data to obtain the probability of assigning individuals to the correct population using each panel. All 96-SNP panels performed similarly and better than the 48-SNP panels except for the 96-SNP BELS panel. Among the 48-SNP panels, panels created from F(ST), I(n), and LC ranks performed better than panels formed using the top-ranked loci from the programs BELS and WHICHLOCI. The application of ranking methods to optimize panel performance will become more important as more high-throughput assays become available.