Motivation: Genome resequencing and short read mapping are two of the primary tools of genomics and are used for many important applications. The current state-of-the-art in mapping uses the quality values and mapping quality scores to evaluate the reliability of the mapping. These attributes, however, are assigned to individual reads and don't directly measure the problematic repeats across the genome. Here we present the Genome Mappability Score (GMS) as a novel measure of the complexity of resequencing a genome. The GMS is a weighted probability that any read could be unambiguously mapped to a given position, and thus measures the overall composition of the genome itself.
Results: We have developed the Genome Mappability Analyzer (GMA) to compute the GMS of every position in a genome. It leverages the parallelism of cloud computing to analyze large genomes, and enabled us to identify the 5-14% of the human, mouse, fly, and yeast genomes that are difficult to analyze with short reads. We examined the accuracy of the widely used BWA/SAMtools polymorphism discovery pipeline in the context of the GMS, and found discovery errors are dominated by false negatives, especially in regions with poor GMS. These errors are fundamental to the mapping process and cannot be overcome by increasing coverage. As such, the GMS should be considered in every resequencing project to pinpoint the dark matter of the genome, including of known clinically relevant variations in these regions.