The most common indel detection methods are alignment-based methods that are generally optimized to detect small indels. These methods are often included in popular variant detection packages such as SAMtools, the genome analysis toolkit (GATK), or VarScan (54–56). Alignment-based methods generally rely on probabilistic models to make indel calls based on data obtained during the initial read mapping and alignment process (Figure 3A). For the example of a small insertion, reads containing the insertion are first mapped to the reference sequence using gapped alignments, a step generally performed by the read mapping software (BWA, Novoalign, etc.) (57). Indel variant detection software will then use the alignment data to call an indel event after applying a filtering step to differentiate common sequence alignment errors from true indels. There are numerous indel detection programs that rely on this method, including Dindel, Stampy, and others, in addition to the more general packages described previously (27,58–61). These methods differ principally in the model used to discriminate between alignment errors and true indel calls, often resulting in greatly discrepant indel calls between software and orthogonally
