When wanting to understand the way a graph G is structured and how the relations it models organize groups of entities, clustering and visualization can be combined to provide the user with a global overview of the graph, on the form of a projected graph: a simplified graph is visualized in which the nodes correspond to a cluster of nodes in the original graph G (with a size proportional to the number of nodes that are classified inside this cluster) and the edges between two nodes have a width proportional to the number of links between the nodes of G classified in the two corresponding clusters. This approach can be trickier when additional attributes (numerical or factors) describe the nodes of G or when the edges of G are of different types and should be treated separately: the simplified representation should then represent similarities for all sets of information. In this proposal, we present a variant of Self-Organizing Maps (SOM), which is adapted to data described by one or several (dis)similarities or kernels recently published in (Olteanu & Villa-Vialaneix, 2015) and which is able to combine clustering and visualization for this kind of graphs.