We recently applied TDA to get more information about the 'relevance' of nodes in graphs, which in turn we employed to improve graph classification. Let me know whether you want to discuss this some more!