I have a large protein-protein interaction based on orthology with experimental PPIs. I want to transform my network to signalling pathways. The NetSearch algorithm can be applied for this but I am looking for much robust solution/software for this.
In order to assess the direction of your network interactions you need to apply an appropriate experimental design to assess the causality. Ideally, you would test the effect of inactivation of your proteins of interest on the rest of the network.
If that is not possible, try using a time-series experiment and a network-building method based on time-delay. The logic of time delay correlation is that the cause always precedes the effect. As such, you can detect the direction of the interactions by using the time of sampling (if the sampling resolution is high enough).
This method is implemented (among others) in the freely available R package TD-ARACNE (check attached link).
I would also like to find software to handle this problem well - I'm assuming that you plan to start with the predicted PPI network and map it to known signalling pathways rather than defining new pathways. There are a few Cytoscape plugins which kind of do what you're talking about. There's the Pathway Scoring Application (http://apps.cytoscape.org/apps/pathwayscoringapplication) and NetAtlas (http://apps.cytoscape.org/apps/netatlas), though depending on the availability of expression data, neither option may provide what you're looking for. You could also just use (or reverse-engineer) the KEGG mapper (http://www.genome.jp/kegg/tool/map_pathway2.html) if you have unique protein IDs.