Figure 3.

Self-comparison approaches. These methods attempt an all-vs.-all self-alignment using the whole assembly or a portion thereof. The self-alignments, viewed as a dot plot, will have many off-diagonal alignments representing dispersed similarities. These methods group the alignments into “piles”, defined by their distinct coverage across a region of the assembly. The primary difference between methods is in how they group piles into families. PILER and CARP require that elements are globally alignable, thereby identifying R1/R3 as a distinct family rather than fragments. Grouper and RECON apply single-linkage clustering, which, in this example, groups all fragments into a single family. RECON further attempts to identify composite families by looking for overrepresented internal edges—in this example, the internal edges were not deemed significant (red x’s).