Sparse Normalized Local Alignment

Given two strings, X and Y, both of length O(n) over alphabet Σ, a basic problem (local alignment) is to find pairs of similar substrings, one from X and one from Y. For substrings X′ and Y′ from X and Y, respectively, the metric we use to measure their similarity is normalized alignment value: LCS(X′,Y′)/(|X′|+|Y′|). Given an integer M we consider only those substrings whose LCS length is at least M. We present an algorithm that reports the pairs of substrings with the highest normalized alignment value in O(nlog|Σ| + rMloglogn) time (r– the number of matches between X and Y). We also present an O(nlog|Σ| + rLloglogn) algorithm (L = LCS(X,Y)) that reports all substring pairs with a normalized alignment value above a given threshold.