em/local_fitting.py

Shows how to locally refine a fit of a protein inside its density using a MC/CG optimization protocol. This example does not necessarily converge to the global minimum as that may require more optimization steps. If one wishes to use this example as a template for real refinement purposes, please adjust the parameters of the function IMP.em.local_rigid_fitting accordingly.

## \example em/local_fitting.py
# Shows how to locally refine a fit of a protein inside
# its density using a MC/CG optimization protocol.
# This example does not necessarily converge to the global minimum
# as that may require more optimization steps.
# If one wishes to use this example as a template for real refinement purposes,
# please adjust the parameters of the function IMP.em.local_rigid_fitting
# accordingly.

import IMP.em
import IMP.core
import IMP.atom
import random
import math
import sys

IMP.setup_from_argv(sys.argv, "local fitting")

IMP.set_log_level(IMP.SILENT)
IMP.set_check_level(IMP.NONE)
m = IMP.Model()
# 1. setup the input protein
# 1.1 select a selector.
# using NonWater selector is more accurate but slower
# sel=IMP.atom.NonWaterPDBSelector()
sel = IMP.atom.CAlphaPDBSelector()
# 1.2 read the protein
mh = IMP.atom.read_pdb(IMP.em.get_example_path("input.pdb"), m, sel)
mh_ref = IMP.atom.read_pdb(IMP.em.get_example_path("input.pdb"), m, sel)
# 1.3 add radius info to each atom, otherwise the resampling would fail.
IMP.atom.add_radii(mh)
IMP.atom.add_radii(mh_ref)
ps = IMP.core.get_leaves(mh)
ps_ref = IMP.core.get_leaves(mh_ref)
# 2. read the density map of the protein
resolution = 8.
voxel_size = 1.5
dmap = IMP.em.read_map(
    IMP.em.get_example_path("input.mrc"), IMP.em.MRCReaderWriter())
dmap.get_header_writable().set_resolution(resolution)
# 3. The protein is now fitted correctly in the density. We can validate
# that by making sure that the cross-correlation score is close to 1.

# 3.1 generate a sampled density map to the same resolution and spacing as
# the target density map. Note that the function we are going to use
# (cross_correlation_coefficient) expects to get the same map dimensions as
# the target density map.
sampled_input_density = IMP.em.SampledDensityMap(dmap.get_header())
sampled_input_density.set_particles(ps)
sampled_input_density.resample()
sampled_input_density.calcRMS()
IMP.em.write_map(sampled_input_density, "vv0.mrc", IMP.em.MRCReaderWriter())
# 3.2 calculate the cross-correlation score, which should be close to 1
best_score = IMP.em.get_coarse_cc_coefficient(
    dmap, sampled_input_density, sampled_input_density.get_header().dmin)
print("The CC score of the native transformation is:", best_score)

# 4. To demonstrate local fitting we locally rotate and translate the
# protein and show how we can go back to the correct placement.

# 4.1 define a local transformation
translation = IMP.algebra.get_random_vector_in(
    IMP.algebra.get_unit_bounding_box_3d())
axis = IMP.algebra.get_random_vector_on(IMP.algebra.get_unit_sphere_3d())
rand_angle = random.uniform(-70. / 180 * math.pi, 70. / 180 * math.pi)
r = IMP.algebra.get_rotation_about_axis(axis, rand_angle)
local_trans = IMP.algebra.Transformation3D(r, translation)
# 4.2 rotate the protein
# prot_xyz=IMP.core.XYZs(IMP.core.get_leaves(mh))
# for xyz in prot_xyz:
#     xyz.set_coordinates(local_trans.get_transformed(xyz.get_coordinates()))
# 4.2 set the protein as a rigid body
IMP.atom.create_rigid_body(mh)
prot_rb = IMP.core.RigidMember(IMP.core.get_leaves(mh)[0]).get_rigid_body()
# 4.3 apply the transformation to the protein
IMP.core.transform(prot_rb, local_trans)
m.update()  # to make sure the transformation was applied
# 4.4 print the new correlation score, should be lower than before
print(len(IMP.core.get_leaves(mh)))
IMP.atom.write_pdb(mh, "input2.pdb")
sampled_input_density.resample()
sampled_input_density.calcRMS()
IMP.em.write_map(sampled_input_density, "vv.mrc", IMP.em.MRCReaderWriter())
start_score = IMP.em.get_coarse_cc_coefficient(
    dmap, sampled_input_density, sampled_input_density.get_header().dmin)
start_rmsd = IMP.atom.get_rmsd(IMP.core.XYZs(ps), IMP.core.XYZs(ps_ref))
print("The start score is:", start_score, "with rmsd of:", start_rmsd)
# 5. apply local fitting
# 5.1 run local fitting
print("performing local refinement, may run for 3-4 minutes")
# translate the molecule to the center of the density
IMP.core.transform(
    prot_rb,
    IMP.algebra.Transformation3D(
        IMP.algebra.get_identity_rotation_3d(),
        dmap.get_centroid() - IMP.core.get_centroid(ps)))
m.update()  # to make sure the transformation was applied
sampled_input_density.resample()
sampled_input_density.calcRMS()
rmsd = IMP.atom.get_rmsd(IMP.core.XYZs(ps), IMP.core.XYZs(ps_ref))
score2 = IMP.em.get_coarse_cc_coefficient(
    dmap, sampled_input_density, sampled_input_density.get_header().dmin)
print("The score after centering is:", score2, "with rmsd of:", rmsd)

refiner = IMP.core.LeavesRefiner(IMP.atom.Hierarchy.get_traits())
fitting_sols = IMP.em.local_rigid_fitting(
    mh.get_particle(), refiner,
    IMP.atom.Mass.get_mass_key(),
    dmap, [], 2, 10, 10)

# 5.2 report best result
# 5.2.1 transform the protein to the preferred transformation
print("The start score is:", start_score, "with rmsd of:", start_rmsd)
for i in range(fitting_sols.get_number_of_solutions()):
    IMP.core.transform(prot_rb, fitting_sols.get_transformation(i))
    m.update()  # to make sure the transformation was applied
    # 5.2.2 calc rmsd to native configuration
    rmsd = IMP.atom.get_rmsd(
        IMP.core.XYZs(ps), IMP.core.XYZs(IMP.core.get_leaves(mh_ref)))
    IMP.atom.write_pdb(mh, "temp_" + str(i) + ".pdb")
    print("Fit with index:", i, " with cc: ", 1. - fitting_sols.get_score(i),
          " and rmsd to native of:", rmsd)
    IMP.atom.write_pdb(mh, "sol_" + str(i) + ".pdb")
    IMP.core.transform(
        prot_rb, fitting_sols.get_transformation(i).get_inverse())
print("done")