The work described in this thesis is part of the analysis of biological phenomena using computers, id est bioinformatics. More precisely, we are interested in nucleic sequence analysis. In this context, our work is splitted in two parts: identification of coding sequences and identification of non-coding sequences that share a common structure such as non-coding RNAs. The main feature of our methods, protea and carnac, is to deal with poorly conserved sequences without the need to align them. Our methods rely on the same comparative analysis scheme to detect evolutionary patterns that are globally coherent between all sequences. \protea and \carnac have been submitted on several reference benchmarks and have reached significative results. We also present two collaborative projects that involve protea and carnac. magnolia is multiple alignement software designed to align nucleic sequences according to their conserved function predicted by protea and/or carnac. The second collaborative project is a software pipeline to automatically annotate genomes by comparative genomics.
