A Bayesian framework for SNP identification

As evolutionary models for single-nucleotide polymorphisms (SNPs) become available, methods for using them in the context of evolutionary information and expert prior information is a necessity. We formulate a probability model for SNPs as a Bayesian inference problem. Using this framework we compare the individual and combined predictive ability of four evolutionary models of varying levels of specificity on three SNP databases (two specifically targeted at functional SNPs) by calculating posterior probabilities and generating Receiver Operating Characteristic (ROC) curves. We discover that none of the models do exceptionally well, in some cases no better than a random-guess model. However, we demonstrate that several properties of the Bayesian formulation improve the predictability of SNPs in the three databases, specifically the ability to utilize mixtures of evolutionary models and a prior based on the genetic code.