Recherche de similarités dans les séquences d'ADN : modèles et algorithmes pour la conception de graines efficaces

Most commonly used similarity search methods in genomic sequences are heuristic ones.These are based upon text ltering that allows one to infer potential regions of similarity. Thisthesis proposes new lter de nitions to search for similarities in genomic sequences, and fastalgorithms to measure the e ciency of these lters.More precisely, we study the spaced seed model and propose an algorithm to measure the seede ciency on similarities of a certain kind, called homogeneous similarities. A generic algorithmhas also been developed to measure the seed e ciency, together with an extension of the spacedseed model called subset seed. Finally, we propose and analyze a multi-seed approach in theframework of lossless ltration, and apply it to the problem of oligonucleotide design.Several software tools have been developed to search for similarities as well as to designseed-based lters.