3D var clustering (in the 3D var display job) is really straightforward right now - it loads in the reaction coordinates (position of each particle along each mode) computed by 3D Var, and then fits a simple mixture of gaussians to the coordinates (in all dimensions simultaneously). The number of gaussians in the mixture is equal to the number of clusters you specify.
Then the particles from each cluster are reconstructed separately and outputted.
You can select which components you want to use for display/clustering by changing the inputs to the 3D var display job. You have to use the “low level results” view in the job builder and the “outputs” tab from the 3D var job.
From the particles, first disconnect all the “components_mode_X” inputs (except components_mode_0 which is required). Then from the 3D var job outputs tab, find the particles components_mode_X outputs that you do want to consider, and start connecting those instead. The first one you connect should be used to override the “compontents_mode_0” input.
Repeat the same process for the volume input (making sure to use the same component numbers for particles and volume).
Unfortunately you will have to consider at least 2 modes (rather than just one) for now, otherwise the display job will fail.
Other than the above, it is still always possible to read out and use the reaction coords from any dimension directly by looking at the .cs file output from 3D var. In that way you could create clusters in any strategy that makes sense.
One thing that I’ve found so far in limited experiments is that completely discrete separable clusters only really appear with relatively large discrete changes in a protein conformation, usually association/dissociation of subunits or really large discrete changes in position of some mass. In more common cases with smaller proteins or with more continuous motion, the distribution of particles along 3D var modes is pretty much continuous. So it might turn out that the simple clustering that is currently being done isn’t the best possible approach - but this is all still work in progress!
Let me know if the above steps make sense!