Classes empiétantes dans un graphe et application aux interactions entre protéines

In this paper, we study a method of classification by density in an unweighted graph. We search some areas with a high density of edges, that can be overlapping (we don't try to obtain a partition but some intrinsic classes). The method consists of two steps ; first we determine the cores of the classes by means of a local density function and then we extend these cores by their neighbourhoods following a criterion on the density of the classes. Finally, the method is applied on a protein-protein interaction network, with the aim of predicting unknown cellular functions of some proteins.