Performing local similarity searches with variable length seeds

Abstract

This paper describes a general method for controlling the running time of similarity search algorithms. Our method can be used in conjunction with the seed-and-extend paradigm employed by many search algorithms, including BLAST. We introduce the concept of a seed tree, and provide a seed tree-pruning algorithm that affects the specificity in a predictable manner. The algorithm uses a single parameter to control the speed of the similarity search. The parameter enables us to reach arbitrary levels between the exponential increases in running time that are typical of seed-and-extend methods. © Springer-Verlag 2004.