Repseek, a tool to retrieve approximate from large DNA sequences

The importance of genome redundancy has been strongly emphasized in the field of genome dynamics and evolution as well as in medical biology. A repeat is a sequence present twice or more with a high degree of similarity within a larger sequence (e.g. a chromosome) or set of sequences (e.g. a genome with several chromosomes). Each instance of the repeated sub-sequence is called a ’copy’ of the repeat. We use the term ”duplication” to denote any active mechanistic event that creates a repeat. Even though spurious duplication events (or recombination events between repeats) can cause severe disorders [26, 24], repeated elements remain nonetheless a very important driving force of genome evolution [28]. In that respect, the dynamics and the evolution of these redundant sequences have been studied in bacterial genomes [31, 32, 5] as well as in eukaryote genomes [3, 4, 38]. Duplication events can sometimes copy entire coding regions, giving birth to what is often referred as duplicate genes. Those duplicate genes are the raw material leading to the emergence of novel functions and have been extensively studied (for a historical review see [37]). Although the repeats we are interested in encompass a lot of known biological repeated elements (i.e. transposable elements, duplicated genes, DNA-satellites, segmental duplication, etc.) our main concern is not to identify specific families of repeats, but to extract repeats on the sole basis of their sequence similarity and without any prior consideration of their biological function. Unlike RepeatMasker [34], we do not search for already well characterized repeated elements. Furthermore, our primary goal is not to construct families of repeats. This is the objective of dedicated software such as RepeatScout [30] or of clustering algorithms [9, 29], which reconstruct families from pairs of repeats. Of course, our program can be used to feed these clustering algorithms. While there are some widely accepted methods to detect duplicate genes in a genome (for instance based on BLAST or FASTA programs), there is no firmly established technique concerning the detection of repeats in large DNA sequences. The detection of repeats is not a trivial problem and there is no satisfactory methodology available apart from recursive local alignment (using dynamic programming) of sequences with themselves [41]. Such algorithms, however, are quadratic in computation time and in memory usage and