We updated the Erl Wood nodes yesterday and the new version is available on the nightlybuilds update site. Beside some bugfixes, the main novelty is a new similarity network viewer in the GPL plugin. It is very similar to the activity cliffs viewer that is contained in the BSD licensed collection of our nodes, but does not require proprietary licenses. Moreover, it is designed to work on arbitrary data type, whereas the activity cliff viewer is restricted to molecular data to offer its full functionality.
I think that similarity network viewer is a good tools expanded not only for compounds. the problem is to obtain a NxN pairwaise distances matrix. Not in knime do that , is it possible to add as calculation node in erl wood group a pairwaise ditance matrix calculator with tanimoto, ecluidean and other metrics?
we also don't have a dedicated node for that. The "Distance Matrix Calculate" node from the Knime Distance Matrix extension does exactly this. It definitely works with RDKit and MOE fingerprints and Tanimoto similarities.
you just have to connect the table containing the similarity matrix to the second input port. We will try to make that clearer in the node description.
I did every combinations, inverting the imput port miss distances matrix and dosen't work. I don't know wich is the problem ...maybe my java version. Could you post a simple workflow as mine that in your hands works? in this case if i receive some error post you the message...and the intere debug.
I modifed to workflow you sent earlier and attached a working version. The top input ports expects a table containing a string id column and an arbitrary data column, the second input port expects a distance matrix column.
thank you, sorry for my errors. This seems as SARANEA software , i use to use the landscape_gui in R as plugin for this , and that are able to do a surface with color coded by activity/property. I think thant your job is great for start to focus on continuos/discontinous omo/etergeneous SAR inside a platform as Knime.
The animation is done using the prefuse library and is motivated by particle dynamics. The reason for that is mainly to layout and spread the network graph, because it is a non-trivial task to draw a complex graph such that is it usable by humans. We usually stop the animation after a certain graph relaxation has occured.
The nodes are coloured using a green -> red heatmap scheme, the edge colouring only highlights activity cliffs. The edge cutoff in the dialog defines the maximum distance in which points are regarded to be neighbors and connected by an edge. The edge colour is controlled by the cliff cutoff value, which is not very intuitive and probably has to be selected by trying out some values.
Generally, the node is designed to be a visualization and not a cliff detection/analyzation - this has to be done by the user.
I started using the network viewer on a data set of roughly about 2000 molecules. Which time period do you recommend for equilibration of the visualisation? It appears to me that I end up with different networks when running the viewer two times in a row - but I'm not sure if the time period was long enough to place the molecules accordingly.
Furthermore, in some cases, the view becomes blank when zooming in - even when the animation is stop. Restarting the animation does not give a new view.
the relaxation is randomized and will lead to different network layout every run. The time until equilibrium depends strongly on the network (in some cases the network will start oscillating and never reach a stable configuration), thus I am afraid that I cannot provide any guidelines.
Regarding the blank visualization bug, I also noticed this behaviour, but struggled with reproducing it and therefore could not really figure the underlying problem. If you recognize a pattern in this behaviour it would be great if you could report it here.
Thanks a lot for the comments and the error reports!
