Mapping short reads, faithfully

Mapping is the process of finding the original location of a DNA read in a reference sequence, typically a genome. Short read mappers are software tools used in most applications that involve high-throughput sequencing. As such, they must be continuously improved to keep up with increasing needs. Modern mappers rely on seeding heuristics, making them fast but inexact. For lack of a method to compute the reliability of their own output, mappers have so far used approximations of variable quality. Here we focus on faithfulness, the capacity to provide accurate mapping confidence, and we devise a strategy to map short reads faithfully. The key is to estimate the repetitiveness of the target reference, which is the dominant factor for the reliability of the mapping process. This approach highlights the existence of a class of reads that can be mapped with unprecedented confidence. We exploit this strategy in a prototype mapper that is competitive with state-of-the-art mappers BWA-MEM and Bowtie2, with the benefit of faithfulness. The software is open-source and available for download at https://github.com/gui11aume/mmp .