JamesK directional PRS 2

[jamesmkrieger]

I've now redone directional PRS, still with plenty of involvement from Claude but starting it myself and putting things in the right place in the code so everything is unified. There is also a new function comparePerturbResponses that allows overlap comparison to ANM modes and deformation vectors, so that most important residues can be identified more easily.

[jamesmkrieger]

For the glutamate receptor ligand-binding domain closure from open dimer from 1fto to glutamate-bound closed dimer AC from 1ftj, we see top contributions from a sequence around L138 that is part of glutamate binding and around K250 in the vicinity of the flip/flop alternative splicing casette that is also related to glutamate response kinetics. So, I think this validates that it captures something useful.

[karolamik13]

I tested the code using the code below. As you can see, I used CA atoms of those structures because they were not equal. Something was missing in one somewhere in the side chain. That is why the results might be different. Btw, structures are aligned via VMD (AB and AC chains).

from prody import *
open_dimer = parsePDB('1fto_aligned.pdb').select('protein and chain A B and resid 5 to 260 and name CA')
closed_dimer = parsePDB('1ftj_aligned.pdb').select('protein and chain A C and resid 5 to 260 and name CA')

target = calcDeformVector(open_dimer, closed_dimer)
anm = ANM('1FTO open')
anm.buildHessian(open_dimer)
anm.calcModes()

responses = calcPerturbResponse(anm, atoms=open_dimer, return_vectors=True, repeats=100)
p = comparePerturbResponses(responses, target, stat='max')

import numpy as np

print(' ')
print('Target vector length:', np.linalg.norm(target.getArray()))
print('RMSD open-closed:', calcRMSD(open_dimer, closed_dimer))

indices = np.argsort(p)[::-1][:20]

print('Chain Residue Score')
for i in indices:
print(f'{open_dimer.getChids()[i]:>5} {open_dimer.getResnames()[i]}{open_dimer.getResnums()[i]:<4} {p[i]:.3f}')

Outcome:

@> 3871 atoms and 1 coordinate set(s) were parsed in 0.04s.
@> 5903 atoms and 1 coordinate set(s) were parsed in 0.06s.
@> Hessian was built in 0.23s.
@> 20 modes were calculated in 0.17s.
@> Calculating perturbation response with 100 repeats
@> Calculating perturbation response [ 99%] 1s
Target vector length: 74.39254136000463
RMSD open-closed: 3.287716904084999
Chain Residue Score
A GLN130 0.631
A LEU127 0.630
A ASP126 0.630
A SER164 0.629
A MET162 0.629
A LYS129 0.627
A ARG163 0.626
A TYR161 0.625
A ALA124 0.625
A SER128 0.625
A THR131 0.625
A SER123 0.624
A GLU125 0.622
B ARG163 0.622
B TRP159 0.622
A ALA165 0.620
B MET158 0.620
B LEU215 0.619
B SER164 0.619
A GLU122 0.619

[jamesmkrieger]

This matches what I saw, thanks. I also used CA atoms and did the matching+alignment with alignChains+superpose in prody.