I. MOTIVATION: Due to the massive amounts of data generated from each instrument run, next generation sequencing technologies have presented researchers with unique analytical challenges which require innovative, computationally efficient statistical solutions. Here we present a parallel implementation of a probabilistic Pair-Hidden Markov Model for base calling and SNP detection in next generation sequencing data. Our approach incorporates multiple sources of error into the base calling procedure which leads to more accurate results. In addition, our approach applies a likelihood ratio test that provides researchers with straight-forward SNP calling cutoffs based on a p-value cutoff or a false discovery control. II. RESULTS: We have developed GNUMAP-SNP, which is a highly accurate approach for the identification of SNPs in next generation sequencing data. By utilizing a novel probabilistic Pair-Hidden Markov Model, GNUMAP-SNP effectively accounts for uncertainty in the read calls as well as read mapping in an unbiased fashion. Our results show that GNUMAP-SNP has both high sensitivity and high specificity throughout the genome, which is especially true in repeat regions or in areas with low read coverage. In addition, we propose a statistical framework that accounts for the background noise using straightforward statistical cutoffs which filters out false-positive results. The parallel implementation of SNP calling achieves near linear speedup on distributed memory or shared memory platforms. III. AVAILABILITY: GNUMAP-SNP is available as a module in the GNUMAP probabilistic read mapping software. GNUMAP is freely available for download at: http://dna.cs.byu.edu/gnumap/.
展开▼
