G-Protein Coupled Receptors (GPCRs) are key players in cell- cell communication. They transduce a wide range of extracellular signals such as light, odors, hormones or neurotransmitters into ap- propriated cellular responses. These receptors regulate many cell functions and are encoded by the largest gene family in mammalian genomes, representing more than 3% of the human genes. GPCRs are the estimated target of approximately half of the medicines cur- rently in clinical use. Probabilistic modelling and specifically, machine learning prob- abilistic models have only recently begun to be applied to the anal- ysis of GPCR functioning, although their application is expected to generate new insights in this field. Statistical machine learning techniques are specially suited to deal with some of the common challenges of molecular modelling in proteins, and should be of spe- cial interest when the three dimensional structures of the proteins and receptors remain unknown at large. In this thesis, we describe a statistical machine learning model of the manifold learning family, adapted through kernelization to the analysis of protein sequence data. Experimental results show that it provides a differentiated visualization and grouping of GPCR subfamilies and that these groupings faithfully reflect the structure of GPCR phylogenetic trees. 3