pyRMSD's pairwiseRMSDMatrix not taking ambiguity into account about imp-sampcon

[jkosinski]

I think I introduced the selection after Ilan's email, so its weird he got different values. I added the selection to narrow down the issue to the Elp2 subunit, but that script is not that important anymore with the new example i posted.

…

On Mon, Aug 2, 2021, 12:20 PM Shruthi Viswanath ***@***.***> wrote: With the previous version of Jan's script pyRMSD_bug.zip <https://github.com/salilab/imp-sampcon/files/6916182/pyRMSD_bug.zip> I could reproduce the numbers 2 A and 114 A. Looks like the script changed to include a selection. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#17 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAMDPOOFXAWSDDTVT3DWNYLT2ZWQRANCNFSM5BDVXH6Q> .

[jkosinski]

@ichem001 if it helps:
I seem to get ambiguity taken into account by the imp-sampcon script if in get_rmsds_matrix of rmsd_calculation.py I replace:

    rmsd = calculator.pairwiseRMSDMatrix()

with

    rmsd = []
    for i in range(len(conforms)-1):
        rmsd.extend(calculator.oneVsFollowing(i))

[ichem001]

@jkosinski

I have tried to reproduce your error but I am not able to reproduce either of the values.

1. Produce the smallest possible example that shows the error. It does not make sense to debug on a six subunits complex.
2. Could you run the test_rmsd_calculation.py test and make sure it passes. That test also makes sure you are using the latest version of imp-sampcom.
3. I cannot reproduce your numbers. I get a RMSD of 76 Å, not of 114 Å.

Let's take it from here to try to understand what the issue is.

[jkosinski]

Hi Ilan,

Thanks for looking into this!

1. Sure, taking your usual structure as an example, building a dimer, and saving two RMF files with identical coordinates but with swapped chains (RMSD should be 0.0 when ambiguity works):

import IMP
import IMP.atom
import IMP.rmf
import IMP.pmi
import IMP.pmi.topology
import RMF
import sys


random_tr3d = IMP.algebra.get_random_local_transformation([0,0,0], 10, 10)


#create two swapped files
for file_i in (0, 1):
    seqs = IMP.pmi.topology.Sequences(IMP.pmi.get_example_path('data/1WCM.fasta'))
    mdl = IMP.Model()
    s = IMP.pmi.topology.System(mdl)
    st = s.create_state()

    #add t
    mol = st.create_molecule("Rpb4", seqs["1WCM:D"], chain_id="D")

    atomic = mol.add_structure(IMP.pmi.get_example_path('data/1WCM_fitted.pdb'),
                        chain_id="D")

    mol.add_representation(
     atomic,
     resolutions=[1])

    clone = mol.create_clone("E")

    hier = s.build()

    if file_i == 0:
        #for the first file, transform the copy 0
        IMP.atom.transform(mol.get_hierarchy(), random_tr3d)
    else:
        #for the second file, transform the copy 1
        IMP.atom.transform(clone.get_hierarchy(), random_tr3d)

    orh = RMF.create_rmf_file('{0}.rmf3'.format(file_i))
    IMP.rmf.add_hierarchy(orh, s.get_hierarchy())
    IMP.rmf.save_frame(orh)
    del orh


#calculate RMSDs

import pyRMSD
from pyRMSD.matrixHandler import MatrixHandler
import scipy
import scipy.spatial
from IMP.sampcon import rmsd_calculation

symmetry = "symm_groups.txt"
with open(symmetry, 'w') as f:
    f.write('Rpb4.0 Rpb4.1')

###########################################
#  reproducing what imp-exhaust is doing: #
###########################################

(ps, masses, radii,
 conforms, symm_groups, models_name,
 n_models) = rmsd_calculation.get_rmfs_coordinates_one_rmf(
     "./",
     '0.rmf3',
     '1.rmf3',
     None,
     symmetry)

rmsd_calculation.get_rmsds_matrix(  # noqa
    conforms, 'cpu_omp', False, 2, symm_groups)

mHandler = MatrixHandler()
mHandler.loadMatrix("Distances_Matrix.data")

rmsd_matrix = mHandler.getMatrix()
distmat = rmsd_matrix.get_data()

distmat_full = scipy.spatial.distance.squareform(distmat)
print('RMSD from get_rmsds_matrix', distmat_full[0,1]) #this should be 0.0 Angstrom but is not !

###############################
# let's try differently then: #
###############################

calculator = pyRMSD.RMSDCalculator.RMSDCalculator(
  'NOSUP_OMP_CALCULATOR',
  fittingCoordsets=conforms,
  calcSymmetryGroups=symm_groups,
  fitSymmetryGroups=symm_groups)

rmsd = calculator.oneVsFollowing(0)[0]
print('RMSD from calculator.oneVsFollowing(0)', rmsd) # now it works, is 0.0 Angstrom !

#but rmsd from pairwiseRMSDMatrix() is still wrong!:
print('RMSD from calculator.pairwiseRMSDMatrix',calculator.pairwiseRMSDMatrix()[0])

###############################
# let's try without ambiguity: #
###############################
calculator = pyRMSD.RMSDCalculator.RMSDCalculator(
  'NOSUP_OMP_CALCULATOR',
  fittingCoordsets=conforms)

rmsd = calculator.oneVsFollowing(0)[0]
print('RMSD from calculator.oneVsFollowing(0) WITHOUT AMBUIGUITY', rmsd) # without ambiguity, the same value as from get_rmsds_matrix!

#but rmsd from pairwiseRMSDMatrix() is still wrong!:
print('RMSD from calculator.oneVsFollowing(0) WITHOUT AMBUIGUITY', calculator.pairwiseRMSDMatrix()[0]) # same here

2. Yes, passes.
3. Hmm, strange, I think Shruthi said she could reproduce. Still, with ambiguity should be much lower than 76A.

[shruthivis]

With the previous version of Jan's script pyRMSD_bug.zip
I could reproduce the numbers 2 A and 114 A. Looks like the script changed to include a selection.