SVseq: an approach for detecting exact breakpoints of deletions with low-coverage sequence data

Motivation: Structural variation (SV), such as deletion, is an important type of genetic variation and may be associated with diseases. While there are many existing methods for detecting SVs, finding deletions is still challenging with low-coverage short sequence reads. Existing deletion finding methods for sequence reads either use the so-called split reads mapping for detecting deletions with exact breakpoints, or rely on discordant insert sizes to estimate approximate positions of deletions. Neither is completely satisfactory with low-coverage sequence reads. Results: We present SVseq, an efficient two-stage approach, which combines the split reads mapping and discordant insert size analysis. The first stage is split reads mapping based on the Burrows-Wheeler transform (BWT), which finds candidate deletions. Our split reads mapping method allows mismatches and small indels, thus deletions near other small variations can be discovered and reads with sequencing errors can be utilized. The second stage filters the false positives by analyzing discordant insert sizes. SVseq is more accurate than an alternative approach when applying on simulated data and empirical data, and is also much faster.