Gap heuristics and tree construction using gaps

The construction of multiple sequence alignments (MSAs) is a fundamental prob-lem in biology. Yet the problem cannot be solved exactly in most cases, as it is exponential in the number of sequences. It is therefore highly desirable to develop heuristics. It is reasonable that not every heuristic starts from scratch, but uses the expertise that has evolved over the years. We present a set of heuristics that take as input an MSA that was produced from any algorithm and produces an improved MSA. The scoring of the resulting alignment is based on a probabilistic model that we developed [16]. With this scoring function we can determine the upper bound (maximum possible score), so we know when the MSA is optimal. The heuristics use the fact that if the sequences in the MSA are related, there exists an (unknown) evolutionary tree. It treats gaps as a special character as gaps play a special role in both protein structure and evolutionary history in order to improve the alignment. From an optimized MSA gaps can be used to reconstruct evolutionary trees. 1