Split read mapping methods, such as the widely used Pindel program, are capable of identifying medium-sized indels that are often missed by alignment-based indel software (33). These methods, including de novo alignment and others, function by first identifying discordant paired-end reads in which one end maps completely to the reference sequence and the other end does not. The unmapped end of these ”single-end anchored reads” are then clustered or subjected to de novo alignment to determine the exact sequence of an insertion (Figure 3B). Using this approach, insertions longer than the read length can be identified, as such methods do not rely on the initial read mapping step. Split read mapping methods are particularly suited to identify clinically relevant indels, such as the FLT3-ITD. However, these methods are also subject to a higher false-positive rate, as they generally do not use probabilistic models to discriminate between alignment errors and true indel events.
