The main part of the thesis is concerned with large-scale studies of codon usage in completely sequenced genomes. A new compositional analysis scheme is presented, complete with a number of computation and visualisation tools. The thesis addresses the benefits of this very general scheme, named codon profiling, with comparisons to the very similar synonymous codon usage. Codon profiling is applied to the analysis of several domains of interest, with the scope of addressing several questions related to the compositional constraints of coding sequences.

The heterogeneity of codon usage in the coding sequences of each genome was examined and presented, noting the consistency of intra-genomic distributions of codon similarity and atypicality. Such distributions provide the grounds on which to elaborate practical applications that make use of these properties.

A computationally inexpensive methodology was developed to detect Horizontal Gene Transfers (and for the first time to identify donor genomes), exploiting measures of codon similarity and combining a compositional identification approach with a phylogenetic verification process.

The thesis also presents a detailed procedure for the characterisation of coding sequences with atypical codon usages, exemplified in a study conducted on a group of human RNA binding proteins whose codon usage has striking similarity to that of some human infecting retroviruses.

Finally, the concept of codon usage space, the space of all the possible codon usages, is discussed. After calculating the theoretical extension of this space, the part visited by known biological sequences was mapped and its dimensionality computed. The comparison with the results obtained using several algorithms for random generation of codon usages quantifies the constraints imposed on biological sequences and allows the investigation and characterisation of the unexplored regions of the space.