I want to find disease drug relationship by using network based approaches. Up til now I was using individual network properties like Strongly connected components (SCC), Betweeness Centrality, transitivity, network diameter and Miniminum dominating set (MDS) but none of these characteristics are yielding the results I am interested in. As example on which I am trying this idea are known (I know the drug causing the phenotype reversion in Gene expression).

Can anybody devise me some other network topology properties like I mentioned, which can be use to compare networks inferred from Gene Expression data?

Secondly I am thinking about combining mentioned features to train the random forest for classification, is it a good idea? If yes, then do share your your experience.

Thanks