Figure 1

The maximal coverage alignment problem: For a query sequence to which multiple local alignments (through standard alignment methods or database search) have been generated, we wish to pick the combination of local alignments that covers the maximum proportion of the query sequence. A) Representing a series of alignments as a graph. Each lettered alignment is shown as a node. Letters indicate the order of these sequences in the reference sequence (i.e. the fourth alignment is letter "a" because it occurs first in the reference sequence). Edges join nodes with permissible overlap. The path through this graph corresponding to the OAC is shown with darkened arrows. At right is a representation of some of the data stored by each node and edge. The five most important pieces of information stored in the nodes are the starting position and length of the alignment (start and len), the alignment score (if that is used as an objective function), the sequence alignment for the node in question (needed when overlap is permitted and alignment scores are the objective function) and the list of "in nodes:" those nodes with a directed edge leading to the current node. Edges store the extent of overlap between the nodes they connect (which is also easily calculated from the two nodes' values of start and len. B) Pseudo-code implementation of the last step in our algorithm, the depth-first search (see text). The dot (.) operator represents access to data structure members.