Parametric Recomuting in Alignment Graphs

DNA/protein sequence alignments in computational molecular biology depend heavily on the settings of penalties for substitutions, insertions/deletions and gaps. Inappropriate choice of parameters causes irrelevant matches (“noise”) to be reported, thus obscuring biologically relevant matches. In practice, biologists frequently compare sequences in a few iterations, starting from a vague idea about appropriate parameters, then refining parameters to reduce noise. This procedure often helps to delineate biologically interesting similarities and to substantially reduce laborious analysis. This paper provides a computational underpinning for such iterative noise filtration in alignment graphs. Our main results assume that a preliminary “noisy” alignment, computed with reasonable but ad hoc parameters, is given; the problem is to modify the parameters to reduce noise. We present fast algorithms to refine penalty parameters and describe an application of these algorithms.