Filtering epitope alignments to improve protein surface prediction

Abstract

In previous work, we developed a new algorithm to computationally predict the epitope, the antibody binding surface of a protein, based on aligning individual mimetic probe sequences derived from an experimental process called antibody imprinting for the protein of interest. A program called EPIMAP implements this algorithm and produces a list of the top-scoring alignment(s) of the probe to protein. Typically 50-100 probes sequences will be known experimentally and must be individually aligned using EPIMAP. The goal of the work reported in this paper is to select the most mutually compatible alignments (one for each probe used) in order to improve the accuracy of epitope prediction. We formalize this problem, show that it is NP-complete and describe an effective branch-and-bound search algorithm that works well in practice for inputs of interest. We show in our experimental results section that filtering alignments improves the accuracy the epitope prediction. © Springer-Verlag 2006.