Shared publicly  - 
Beautiful work from +Iain Wallace - active antimalarials' targets expressed as compound similarity maps. Summary post:

...with links to the primary data therein. #Cytoscape users can enjoy the full experience by downloading the .cys files.

Iain - what this generates in my mind is thoughts about what to make next, for which we'd need easy visualization of structures, but that's possible within cytoscape. However, a short-term question I ask myself is whether we can gain access to these compounds - i.e. whether a commercial supplier is listed for any of them. But then I realise that these compounds are grouped by predicted biological target, meaning we ought to be using that aspect of the data, meaning we need to know if any of these compounds have KNOWN (proven) targets, i.e. the strength of the predicted target for a given compound depends on the aggregated strengths of the predictions of the surrounding compounds - is that right?
Iain Wallace's profile photo
Sure, for the next iteration depending on the feedback, we can add on purchasable information from e-molecules.

I have done this on a small scale for the OpenBox compounds and the fastrack GSK series. They are are uploaded to

Just to clarify, the network is clustered based on compound similarity, so similar compounds are grouped together.
The target prediction is over-layed onto the network using colour to represent predicted target. Similar coloured nodes cluster together, in general, because the prediction is made based on compound structure and the network is clustered based on compound structure. i.e. similar compounds cluster together because of similar structure, and because they have similar structure they often share a predicted target.

I really really would suggest, if you like the visualisation, that you download cytoscape and play around with the network. Cytoscape is a really great tool once you start using it. Wet lab biologists that I work with often do, and if biologists can why not chemists :)

As a matter of interest, what other types of data would you use to decide what compounds to prioritise? I can understand purchasability and predicted targets, but am wondering are we missing anything else. We could add in Lipinsk's/LogP flags too.
Add a comment...