Fig. 1.

Overview of the GCM algorithm. The input is a set of constraint alignments. (i) We compute a set of backbone alignments with the same number of sequences from each constraint alignment. (ii) We construct the alignment graph. Each node represents a column from a constraint alignment, and weighted edges are derived from homologies inferred in the backbone alignments: thicker lines represent edges with higher weight. (iii) We cluster the alignment graph with MCL. This example shows two violations: the pink-outlined cluster contains two columns from the same (blue) constraint alignment, and there is no valid ordering between the pink- and green-outlined clusters (‘crisscrossing’). (iv) We resolve the violations and produce a valid trace, where each connected component is a column in our final alignment