Networks and Their Uses
It sometimes takes me a little bit of time to understand and appreciate concepts that others pick up easily; one of these areas is around pathways and networks. The diagrams are great, and make great visual images. I have one on my office wall. It seems the more connections there are, the better - but can people understand these big messes as opposed to a simplified critical path map? There are now some great tools to explore these networks, things like cytoscape, and the annotation of these pathway diagrams is clearly valuable. One of the best books I’ve read of the broad field was the early book by Barabasi - ‘Linked’, but this is quite old now, and there has been a lot of progress.
I was at an excellent workshop this week at the NIH in Bethesda, and I was fortunate enough to participate in a specific workshop session around networks and perturbagens. This had some of the real thought-leaders of the field in, and it was great to put a face to some of the names from papers I had read. I do have a couple of things I still haven’t worked out yet, so I thought I’d try and write down what they are, and pose some questions to the ChEMBL-oggers for clarification, trigger some discussion, chat about examples, etc.
- Fundamentally I feel that these networks must be most valuable when they are based on physical interactions, these must give better insight into mechanism. However, it is really common to have links between genes on the basis of transcriptional co-regulation, and given that there appears to be no standards for the display of physical compared to virtual interactions, I find working out what on earth is going on with a diagram really complex. Are there standard ways of representing the classes of interaction (physical obligate, physical transient, transcriptional, etc)?
- Are there much literature connected to a domain level annotation of the networks; for example, prothrombin, the gla-domains, the kringles and the serine proteinase domain, etc, they all have different partners, and their partners usually bind at different surfaces. In terms of using the networks to think about regulation, surely the definition or annotation of the eptopes/binding sites must be essential in order to mine these networks?
- How do people deal with mapping the effect of purturbagens onto these networks - the whole of pharmacology and drug discovery is built around quantitation of dose response, and there must be ways of showing the reconfiguration of networks as concentration of a perturbagen is raised?
- The stability/stoichiometry of the components of these networks seems to be typically ignored - homo-oligomerisation is surely a valid class of interaction, and a lot of biology is based around concentration dependent oligomerisation events. Also people typically don’t seem to worry too much about stable complexes, versus more dynamic more highly kinetic interactions. Is this as big an issue as I think it might be?
- How do researchers define the boundaries of these networks, is there an accepted convention to stop these networks growing too large - so how do people do pruning?
- Is there anyway that people validate these networks, a sort of statistic that measures the statistical support for a particular topology, or fragility of the particular rendering of the network.?
- Finally, probably again related to the validation of these networks and their usefulness in quantitative biology - do people routinely map available assays across the nodes, of course these could span many connected nodes, but mapping these pathways to bioassays must be a useful thing? This would clearly be a cool thing to try and do with ChEMBL.