ParMap, an algorithm for the identification of small genomic insertions and deletions in nextgen sequencing data

Here we describe ParMap, a statistical algorithm for the identification of complex genetic variants, such as small insertion and deletions, using partially mapped reads in nextgen sequencing data.We report ParMap's successful application to the mutation analysis of chromosome X exome-captured leukemia DNA samples.One of the major technological advances in biology in the last few years has been the development of high throughput nextgen sequencing systems that produce gigabases of data in a single run, and allow an unbiased view of the whole genome without relying on prior knowledge about the disease-causing alterations. These ultradeep sequencing technologies produce large amounts of sequence data, which increase the sequencing depth and allow for better statistics in calling various genomic variations. However, they do so at the cost of reducing the read length and increasing the error rate relative to traditional Sanger sequencing. Thus, the development of efficient statistical and computational methods for the high confidence call of genomic variants is needed for the analysis of these high throughput datasets.At this point, the detection of single mutations and large copy number variations using deep sequencing data is fairly straight forward [1,2], whereas the identification of small (less than 10 nucleotides) insertions and deletions is more challenging. A few algorithms have been developed for detecting complex genomic variants, such as structural variations and insertions and deletions, using mate-pair or paired-end reads [3,4], however, identifying small insertion/deletions in fragment (single-end) data has proved to be very difficult. Mapping algorithms that are designed for very short reads have to assign large penalties for introducing gaps in the middle of the alignment in order to map the majority of the reads efficiently. Conversely, decreasing the gap penalties, increases the number of reads that are mapped with low confidence. However, these methods