ppillot
commented
1 year ago Hi Pedro. I don't think there is anything built-in that can help you there. Selections represent atoms, but centroids are not.
What you can do is:
- create your selection
- pass it to the structure.eachResidue(callback(residueProxy), selection) function
- in your callback, executed for each residue, create an empty vector 3, use the residueProxy.eachAtom(callback(atomProxy)) iterator.
- in this callback, filter the atoms that are in the sidechain (I think it is atomProxy.isSidechain) and if they are, add their coordinates to the Vector3 object for the residue
- after the callback, divide the Vector3 by the number of atoms in the side chain: this gives you the coordinates of the centroid for this side chain
For the rest of your algorithm, you manipulate only Vector3 objects so distanceTo() is available. If you are only comparing distances, a minor optimization is to use the squared distances instead (avoids a root square operation). There might be some other optimizations in the comparisons (e.g. not doing an all against distance which is O(n2) in time complexity, ~but use the centroid of all centroids (which you can compute in O(n) time just by adding Vector3 objects) and compare the distance to this centroid; I am not certain it will give you the same result though, need to check that~) Scratch that... What you are trying to do is clustering residues by distance. There are methods to do clustering in linear time but this is not straightforward.
pguillem
I'm trying to build a function that:
1) Receives a selection of arbitrary residues (say.. 20 residues) 2) Calculate the centroid of each sidechain (maybe an xyz position in space, or simply a C near the centroid). 3) Measure the distance from all to all centroids. 4) Create groups of residues separated by less than the average distance from all to all.
My questions:
Thanks in advance... I know perhaps this is a very specific thing to do. Pedro