Classification of SNP genotypes by a Gaussian mixture model in competitive enzymatic assays

We present statistical models and methods for classification of bi-allelic SNP genotypes when data represent two signal intensities, one signal x from a primer matching one of the alleles, and the other signal y matching the other allele. One such technique is protease-mediated allele-specific extension (PrASE), and the study is at the same time a case study on PrASE data. Most information for classification is contained in the variate log(x/y), for which we derive a special 3-component mixture model from molecular principles. We describe inference in this mixture model, but we also discuss other topics such as assessing the number of components, the information available in the orthogonal variation, detection of overall outlying individuals, and supplementary use of the Hardy–Weinberg law.