Leaflet clustering over time works for all tested lipid topologies such as stacked bilayers, membrane tethers, complex bilayers, protein containing complex bilayers and inverted haxagonal phases fusing on top of a bilayer. However much has to be tested to guarantee the quality of the clustering and its behaviour. A few quick guidelines if you would like to use this version:
1) Put the clustering_input.py associated to your tpr and xtc and link accordingly.
2) Set the clustering resolution (1nm is fine for most stuff unless you want more detail)
-- Whenever you are going below 0.9 nm resolution for CG systems it is highly advised to used hyper_res. This uses blurring of the individual positions over half the resolution for improved binning behaviour, but could allow for double assignment to clusters in residues of which the last one would overwrite, possibly giving some noice.
3) Turn on force clustering or not. Due to the clustering methodology, lipids which are diving in between lipid tails have a high chance of not being clustered initially. These 'skipped' lipids can be forced to a cluster by checking the most prevalent cluster within a cutoff from their 0 bead (always a relevant headgroup?).
-- If you turn on the force_info you will get information about what is not clustered before and after forcing it. This to see if it makes sense to you.
4) Once you checked this you can run the clustering_input.py with python3
5) This will created a clusters.npy and clusters_ordered.npy, the last one has consistent cluster ID over time.
6) Use the vmd_visualization script (by default it expects a ref.gro and ref.xtc as input files and this should be changed accordingly in the small script itself). You can use the VMD visualization script if you have VMD compiled with python and numpy. To do so start VMD and type 'gopython name_of_the_visualization_script.py'. It will load the trajectory and add the cluster IDs under the custom 'user' selection.
7) Make a cluster selection by typing 'user 1' to see the first cluster. Cluster 0 is the void cluster.
8) You can save the representation, but you will have to always start with the VMD python script and then load your saved visualization on top (you should remove loading the gro and xtc from the VMD visualization state file).
I hope this if informative to allow for some use and if you do use it, please let us know what you think.