FAMOUS: Fast Approximate string Matching using OptimUm search Schemes.

Finding approximate occurrences of a pattern in a text using a full-text index is a central problem in bioinformatics. Bidirectional indices have opened new possibilities for solving the problem as they allow the search to be started from anywhere within the pattern and extended in both directions. In particular, use of search schemes (partitioning the pattern into several pieces and searching the pieces in certain orders with bounds on the number of errors in each piece) has shown significant potential in speeding up approximate matching. However, finding the optimal search scheme to maximize the search speed is a difficult combinatorial optimization problem. In this paper, we propose, for the first time, a method to solve the optimal search scheme problem for Hamming distance with given number of pieces. Our method is based on formulating the problem as a mixed integer program (MIP). We show that the optimal solutions found by our MIP significantly improve upon previously published ad-hoc solutions. Our MIP can solve problems of considerable size to optimality in reasonable time and has the attractive property of finding near-optimal solutions for much larger problems in a very short amount of time. In addition, we present FAMOUS (Fast Approximate string Matching using OptimUm search Schemes), a bidirectional search (for Hamming and edit distance) implemented in SeqAn that performs the search based on the optimal search schemes from our MIP. We show that FAMOUS is up to 35 times faster than standard backtracking and anticipate that it will improve many tools as a new core component for approximate matching and NGS data analysis. We exemplify this by searching Illumina reads completely in our index at a speed comparable to or faster than current read mapping tools. Finally, we pose several open problems regarding our MIP formulation and use of its solutions in bidirectional search.