A New Genotype Calling Method For Affymetrix SNP Arrays

Current genotype calling methods such as RLMM and CRLMM provide accurate calling results for Affymetrix SNP chips. However in the small sample case the accu-racy rate may drop significantly and the calling results are not consistent as the sample size changes. And these methods are computationally expensive as they employ pre-process procedures including chip data normalization and other sophisticated statistical techniques.We develop a new genotype calling method for Affymetrix 100k and 500k SNP chips. A two-stage classification scheme is proposed to obtain a fast genotype calling algorithm. It is found that in the first stage the unsupervised classification can easily discriminate genotypes with high accuracy for more than 50% SNPs. And in the second stage the supervised classification employs the modified Mahalanobis distance which incorporates the allele frequency information from the HapMap training data and results in remarkable improvement over the usual Mahalanobis distance based classifier.The overall performance in term of accuracy rate is shown to be comparable to that of the CRLMM as verified by the known golden standard HapMap data and is superior to the competing ones in small sample cases. Confidence score is provided for every genotype call as well. The new algorithm is computationally simple and standalone in some sense that a self-training scheme can be used without employing any other training data. What is more, the training sets built from the HapMap data are easy to update when new validated genotype information become available. And the self-training process is effective when the size of the test sample is moderate or large.