The diversity of life derives mostly from the variety of proteins coded in genomes. How did evolution produce such a tremendous diversity ? The classical theory postulates that this diversity results both from sequence divergence and from the combinatorial arrangements of a few thousand primary protein domain types. However this does not account for the increasing number of entirely unique proteins as found in most genomes.In this thesis, we study the evolution of proteins from the point of view of their domain decomposition and rely on three databases : HOGENOM (homologous protein families), Pfam (manually curated protein domain families) and ProDom (automatically built protein module families). Each protein family from HOGENOM has thus been decomposed into Pfam domains or ProDom modules. We have modelled the evolution of these families using a Bayesian network based on the phylogenetic species tree. In the framework of this model, we can rigorously reconstitute the most likely evolutionary scenarios reflecting the presence or absence of each protein, domain or module in ancestral species. The comparison of these scenarios allows us to analyse the emergence of new proteins in terms of ancestral domains or modules. Pfam analysis suggests that the majority of protein innovations results from rearrangements of ancient domains, in agreement with the classical paradigm of modular protein evolution. However a very significant part of protein diversity is then neglected. On the other hand ProDom analysis suggests that the majority of new proteins have recruited novel protein modules. We discuss the respective biases of Pfam and ProDom underlying these contrasting views. We propose that the emergence of new protein modules may result from a fast turnover of coding sequences and that this module innovation is essential to the emergence of numerous novel proteins throughout evolution
