Recursive Dynamic Programming for Adaptive Sequence and Structure Alignment

We propose a new alignment procedure that is capable of aligning protein sequences and structures in a unified manner. Recursive dynamic programming (RDP) is a hierarchical method which, on each level of the hierarchy, identifies locally optimal solutions and assembles them into partial alignments of sequences and/or structures. In contrast to classical dynamic programming, RDP can also handle alignment problems that use objective functions not obeying the principle of prefix optimality, e.g. scoring schemes derived from energy potentials of mean force. For such alignment problems, RDP aims at computing solutions that are near-optimal with respect to the involved cost function and biologically meaningful at the same time. Towards this goal, RDP maintains a dynamic balance between different factors governing alignment fitness such as evolutionary relationships and structural preferences. As in the RDP method gaps are not scored explicitly, the problematic assignment of gap cost parameters is circumvented. In order to evaluate the RDP approach we analyse whether known and accepted multiple alignments based on structural information can be reproduced with the RDP method. For this purpose, we consider the family of ferredoxins as our prime example. Our experiments show that, if properly tuned, the RDP method can outperform methods based on classical sequence alignment algorithms as well as methods that take purely structural information into account.