def conjugate_gradient_optimization(model, log_energies):
restraints = []
# IMP COMMAND: Call the Conjugate Gradient optimizer
try:
restraints.append(model['restraints'])
scoring_function = IMP.core.RestraintsScoringFunction(restraints)
except:
pass
try:
lo = IMP.core.ConjugateGradients()
lo.set_model(model['model'])
except: # since version 2.5, IMP goes this way
lo = IMP.core.ConjugateGradients(model['model'])
try:
lo.set_scoring_function(scoring_function) # 2.6.1 compat
except:
pass
#print LSTEPS, NROUNDS, STEPS
o = IMP.core.MonteCarloWithLocalOptimization(lo, LSTEPS)
try:
o.set_scoring_function(scoring_function) # 2.6.1 compat
except:
pass
o.set_return_best(True)
fk = IMP.core.XYZ.get_xyz_keys()
ptmp = model['particles'].get_particles()
x, y, z, radius = (FloatKey("x"), FloatKey("y"), FloatKey("z"),
FloatKey("radius"))
mov = IMP.core.NormalMover(ptmp, fk, 0.25 / SCALE)
o.add_mover(mov)
# o.add_optimizer_state(log)
# Start optimization and save a VRML after 100 MC moves
try:
log_energies.append(model['model'].evaluate(False))
except:
log_energies.append(
model['restraints'].evaluate(False)) # 2.6.1 compat
#"""simulated_annealing: perform simulated annealing for at most nrounds
# iterations. The optimization stops if the score does not change more than
# a value defined by endLoopValue and for stopCount iterations.
# @param endLoopCount = Counter that increments if the score of two models
# did not change more than a value
# @param stopCount = Maximum values of iteration during which the score
# did not change more than a specific value
# @paramendLoopValue = Threshold used to compute the value that defines
# if the endLoopCounter should be incremented or not"""
# IMP.fivec.simulatedannealing.partial_rounds(m, o, nrounds, steps)
endLoopCount = 0
stopCount = 10
endLoopValue = 0.00001
# alpha is a parameter that takes into account the number of particles in
# the model (len(LOCI)).
# The multiplier (in this case is 1.0) is used to give a different weight
# to the number of particles
alpha = 1.0 * len(LOCI)
# During the firsts hightemp iterations, do not stop the optimization
hightemp = int(0.025 * NROUNDS)
for i in range(0, hightemp):
temperature = alpha * (1.1 * NROUNDS - i) / NROUNDS
o.set_kt(temperature)
log_energies.append(o.optimize(STEPS))
#if i == 1:
# for p in ptmp:
# print p,p.get_value(x),p.get_value(y),p.get_value(z)
if VERBOSE == 3:
print(i, log_energies[-1], o.get_kt())
# After the firsts hightemp iterations, stop the optimization if the score
# does not change by more than a value defined by endLoopValue and
# for stopCount iterations
lownrj = log_energies[-1]
for i in range(hightemp, NROUNDS):
temperature = alpha * (1.1 * NROUNDS - i) / NROUNDS
o.set_kt(temperature)
log_energies.append(o.optimize(STEPS))
if VERBOSE == 3:
print(i, log_energies[-1], o.get_kt())
# Calculate the score variation and check if the optimization
# can be stopped or not
if lownrj > 0:
deltaE = fabs((log_energies[-1] - lownrj) / lownrj)
else:
deltaE = log_energies[-1]
if (deltaE < endLoopValue and endLoopCount == stopCount):
break
elif (deltaE < endLoopValue and endLoopCount < stopCount):
endLoopCount += 1
lownrj = log_energies[-1]
else:
endLoopCount = 0
lownrj = log_energies[-1]
def generate_IMPmodel(rand_init):
"""
Generates one IMP model
:param rand_init: random number kept as model key, for reproducibility.
:returns: a model, that is a dictionary with the log of the objective
function value optimization, and the coordinates of each particles.
"""
verbose = VERBOSE
IMP.random_number_generator.seed(rand_init)
log_energies = []
model = {
'rk': IMP.FloatKey("radius"),
'model': Model(),
'ps': None,
'pps': None
}
model['ps'] = ListSingletonContainer(
IMP.core.create_xyzr_particles(model['model'], len(LOCI), RADIUS,
100000))
model['ps'].set_name("")
# initialize each particles
for i in range(0, len(LOCI)):
p = model['ps'].get_particle(i)
p.set_name(str(LOCI[i]))
# computed following the relationship with the 30nm vs 40nm fiber
#p.set_value(model['rk'], RADIUS)
# Restraints between pairs of LOCI proportional to the PDIST
try:
# version: 847e65d44da7d06718bcad366b09264c818752d5
model['pps'] = IMP.ParticlePairs()
except AttributeError:
model['pps'] = IMP.kernel.ParticlePairsTemp()
# CALL BIG FUNCTION
if rand_init == 1 and verbose == 0.5:
verbose = 1
stdout.write("# Harmonic\tpart1\tpart2\tdist\tkforce\n")
# set container
model['container'] = CONFIG['container']
if model['container']['shape'] == 'cylinder':
# define a segment of a given size
segment = IMP.algebra.Segment3D(
IMP.algebra.Vector3D(0, 0, 0),
IMP.algebra.Vector3D(model['container']['height'], 0, 0))
bb = IMP.algebra.get_bounding_box(segment)
rb = IMP.container.SingletonsRestraint(
IMP.core.BoundingBox3DSingletonScore(
IMP.core.HarmonicUpperBound(model['container']['radius'],
model['container']['cforce']), bb),
model['ps'])
model['model'].add_restraint(rb)
rb.evaluate(False)
# elif model['container']['shape']:
# raise noti
addAllHarmonics(model)
# Setup an excluded volume restraint between a bunch of particles
# with radius
r = IMP.core.ExcludedVolumeRestraint(model['ps'], CONFIG['kforce'])
model['model'].add_restraint(r)
if verbose == 3:
print "Total number of restraints: %i" % (
model['model'].get_number_of_restraints())
# Set up optimizer
lo = IMP.core.ConjugateGradients()
lo.set_model(model['model'])
o = IMP.core.MonteCarloWithLocalOptimization(lo, LSTEPS)
o.set_return_best(True)
fk = IMP.core.XYZ.get_xyz_keys()
ptmp = model['ps'].get_particles()
mov = IMP.core.NormalMover(ptmp, fk, 0.25)
o.add_mover(mov)
# o.add_optimizer_state(log)
# Optimizer's parameters
if verbose == 3:
print "nrounds: %i, steps: %i, lsteps: %i" % (NROUNDS, STEPS, LSTEPS)
# Start optimization and save an VRML after 100 MC moves
log_energies.append(model['model'].evaluate(False))
if verbose == 3:
print "Start", log_energies[-1]
#"""simulated_annealing: preform simulated annealing for at most nrounds
# iterations. The optimization stops if the score does not change more than
# a value defined by endLoopValue and for stopCount iterations.
# @param endLoopCount = Counter that increments if the score of two models
# did not change more than a value
# @param stopCount = Maximum values of iteration during which the score
# did not change more than a specific value
# @paramendLoopValue = Threshold used to compute the value that defines
# if the endLoopCounter should be incremented or not"""
# IMP.fivec.simulatedannealing.partial_rounds(m, o, nrounds, steps)
endLoopCount = 0
stopCount = 10
endLoopValue = 0.00001
# alpha is a parameter that takes into account the number of particles in
# the model (len(LOCI)).
# The multiplier (in this case is 1.0) is used to give a different weight
# to the number of particles
alpha = 1.0 * len(LOCI)
# During the firsts hightemp iterations, do not stop the optimization
hightemp = int(0.025 * NROUNDS)
for i in range(0, hightemp):
temperature = alpha * (1.1 * NROUNDS - i) / NROUNDS
o.set_kt(temperature)
log_energies.append(o.optimize(STEPS))
if verbose == 3:
print i, log_energies[-1], o.get_kt()
# After the firsts hightemp iterations, stop the optimization if the score
# does not change by more than a value defined by endLoopValue and
# for stopCount iterations
lownrj = log_energies[-1]
for i in range(hightemp, NROUNDS):
temperature = alpha * (1.1 * NROUNDS - i) / NROUNDS
o.set_kt(temperature)
log_energies.append(o.optimize(STEPS))
if verbose == 3:
print i, log_energies[-1], o.get_kt()
# Calculate the score variation and check if the optimization
# can be stopped or not
if lownrj > 0:
deltaE = fabs((log_energies[-1] - lownrj) / lownrj)
else:
deltaE = log_energies[-1]
if (deltaE < endLoopValue and endLoopCount == stopCount):
break
elif (deltaE < endLoopValue and endLoopCount < stopCount):
endLoopCount += 1
lownrj = log_energies[-1]
else:
endLoopCount = 0
lownrj = log_energies[-1]
#"""simulated_annealing_full: preform simulated annealing for nrounds
# iterations"""
# # IMP.fivec.simulatedannealing.full_rounds(m, o, nrounds, steps)
# alpha = 1.0 * len(LOCI)
# for i in range(0,nrounds):
# temperature = alpha * (1.1 * nrounds - i) / nrounds
# o.set_kt(temperature)
# e = o.optimize(steps)
# print str(i) + " " + str(e) + " " + str(o.get_kt())
log_energies.append(model['model'].evaluate(False))
if verbose >= 1:
if verbose >= 2 or not rand_init % 100:
print 'Model %s IMP Objective Function: %s' % (rand_init,
log_energies[-1])
x, y, z, radius = (FloatKey("x"), FloatKey("y"), FloatKey("z"),
FloatKey("radius"))
result = IMPmodel({
'log_objfun': log_energies,
'objfun': log_energies[-1],
'x': [],
'y': [],
'z': [],
'radius': None,
'cluster': 'Singleton',
'rand_init': str(rand_init)
})
for part in model['ps'].get_particles():
result['x'].append(part.get_value(x))
result['y'].append(part.get_value(y))
result['z'].append(part.get_value(z))
if verbose == 3:
print(part.get_name(), part.get_value(x), part.get_value(y),
part.get_value(z), part.get_value(radius))
# gets radius from last particle, assuming that all are the same
result['radius'] = part.get_value(radius)
return result # rand_init, result
def generate_IMPmodel(rand_init,
HiCRestraints,
use_HiC=True,
use_confining_environment=True,
use_excluded_volume=True,
single_particle_restraints=None):
"""
Generates one IMP model
:param rand_init: random number kept as model key, for reproducibility.
:returns: a model, that is a dictionary with the log of the objective
function value optimization, and the coordinates of each particles.
"""
verbose = VERBOSE
# Set the IMP.random_number_generator.seed to get a different reproducible model
IMP.random_number_generator.seed(rand_init)
#print IMP.random_number_generator()
log_energies = []
model = {
'radius': IMP.FloatKey("radius"),
'model': Model(),
'particles': None,
'restraints': None
} # 2.6.1 compat
model['particles'] = ListSingletonContainer(
model['model'],
IMP.core.create_xyzr_particles(model['model'], len(LOCI), RADIUS,
100000 /
SCALE)) # last number is box size
try:
model['restraints'] = IMP.RestraintSet(model['model']) # 2.6.1 compat
except:
pass
# OPTIONAL:Set the name of each particle
for i in range(0, len(LOCI)):
p = model['particles'].get(i)
p.set_name(str(LOCI[i]))
#print p.get_name()
# Separated function for the confining environment restraint
if use_confining_environment:
# print "\nEnforcing the confining environment restraint"
add_confining_environment(model)
# Separated function for the bending rigidity restraints
if CONFIG['kbending'] > 0.0:
# print "\nEnforcing the bending rigidity restraint"
theta0 = 0.0
bending_kforce = CONFIG['kbending']
add_bending_rigidity_restraint(model, theta0, bending_kforce)
# Add restraints on single particles
if single_particle_restraints:
for ap in single_particle_restraints:
ap[1] = [(c / SCALE) for c in ap[1]]
ap[4] /= SCALE
# This function is specific for IMP
add_single_particle_restraints(model, single_particle_restraints)
# Separated function fot the HiC-based restraints
if use_HiC:
# print "\nEnforcing the HiC-based Restraints"
HiCbasedRestraints = HiCRestraints.get_hicbased_restraints()
add_hicbased_restraints(model, HiCbasedRestraints)
# Separated function for the excluded volume restraint
if use_excluded_volume:
# print "\nEnforcing the excluded_volume_restraints"
excluded_volume_kforce = CONFIG['kforce']
evr = add_excluded_volume_restraint(model, model['particles'],
excluded_volume_kforce)
if verbose == 1:
try:
print("Total number of restraints: %i" %
(model['model'].get_number_of_restraints()))
if use_HiC:
print(len(HiCbasedRestraints))
except:
print("Total number of restraints: %i" %
(model['restraints'].get_number_of_restraints())
) # 2.6.1 compat
if use_HiC:
print(len(HiCbasedRestraints))
# Separated function for the Conjugate gradient optimization
if verbose == 1:
print("\nPerforming the optimization of the Hi-C-based restraints "
"using conjugate gradient optimisation\n")
conjugate_gradient_optimization(model, log_energies)
#try:
# log_energies.append(model['model'].evaluate(False))
#except:
# log_energies.append(model['restraints'].evaluate(False)) # 2.6.1 compat
if verbose >= 1:
if verbose >= 2 or not rand_init % 100:
print('Model %s IMP Objective Function: %s' %
(rand_init, log_energies[-1]))
x, y, z, radius = (FloatKey("x"), FloatKey("y"), FloatKey("z"),
FloatKey("radius"))
result = IMPmodel({
'log_objfun': log_energies,
'objfun': log_energies[-1],
'x': [],
'y': [],
'z': [],
'radius': None,
'cluster': 'Singleton',
'rand_init': str(rand_init)
})
for part in model['particles'].get_particles():
result['x'].append(part.get_value(x) * SCALE)
result['y'].append(part.get_value(y) * SCALE)
result['z'].append(part.get_value(z) * SCALE)
if verbose == 3:
print((part.get_name(), part.get_value(x), part.get_value(y),
part.get_value(z), part.get_value(radius)))
# gets radius from last particle, assuming that all are the same
# include in the loop when radius changes... should be a list then
result['radius'] = part.get_value(radius)
result['radius'] = SCALE / 2
#for log_energy in log_energies:
# print "%.30f" % log_energy
#print rand_init, log_energies[-1]
#stdout.flush()
return result # rand_init, result
def generate_IMPmodel(rand_init,
HiCRestraints,
use_HiC=True,
use_confining_environment=True,
use_excluded_volume=True,
single_particle_restraints=None,
initial_conformation=None):
"""
Generates one IMP model
:param rand_init: random number kept as model key, for reproducibility.
:returns: a model, that is a dictionary with the log of the objective
function value optimization, and the coordinates of each particles.
"""
verbose = VERBOSE
# Set the IMP.random_number_generator.seed to get a different reproducible model
IMP.random_number_generator.seed(rand_init)
log_energies = []
model = {
'radius': IMP.FloatKey("radius"),
'model': Model(),
'particles': None,
'restraints': None
} # 2.6.1 compat
if initial_conformation:
ps = []
for pos in xrange(len(LOCI)):
ps.append(IMP.Particle(model['model'], str(pos)))
d = IMP.core.XYZR.setup_particle(ps[-1], float(RADIUS))
d.set_coordinates(
IMP.algebra.Vector3D(initial_conformation['x'][pos],
initial_conformation['y'][pos],
initial_conformation['z'][pos]))
d.set_coordinates_are_optimized(True)
model['particles'] = ListSingletonContainer(model['model'], ps)
else:
model['particles'] = ListSingletonContainer(
IMP.core.create_xyzr_particles(
model['model'], len(LOCI), RADIUS,
100000 / float(CONFIG['resolution'] *
CONFIG['scale']))) # last number is box size
# OPTIONAL:Set the name of each particle
for i in range(0, len(LOCI)):
p = model['particles'].get_particle(i)
p.set_name(str(LOCI[i]))
#model['particles'] = ListSingletonContainer(
# IMP.core.create_xyzr_particles(model['model'], len(LOCI), RADIUS, 100000)) # last number is box size
try:
model['restraints'] = IMP.RestraintSet(model['model']) # 2.6.1 compat
except:
pass
# Separated function for the confining environment restraint:
# This function is specific for IMP
if use_confining_environment:
# print "\nEnforcing the confining environment restraint"
add_confining_environment(model)
# Separated function for the bending rigidity restraints
# This function is specific for IMP
if CONFIG['kbending'] > 0.0:
# print "\nEnforcing the bending rigidity restraint"
theta0 = 0.0
bending_kforce = CONFIG['kbending']
add_bending_rigidity_restraint(model, theta0, bending_kforce)
# Anchor particles to fixed points in the model
if single_particle_restraints:
for ap in single_particle_restraints:
ap[1] = [
c / (float(CONFIG['resolution'] * CONFIG['scale']))
for c in ap[1]
]
ap[4] /= float(CONFIG['resolution'] * CONFIG['scale'])
# This function is specific for IMP
add_single_particle_restraints(model, single_particle_restraints)
# Separated function fot the HiC-based restraints
if use_HiC:
# This function is general
HiCbasedRestraints = HiCRestraints.get_hicbased_restraints()
# print "\nEnforcing the HiC-based Restraints
# This function is specific for IMP
add_hicbased_restraints(model, HiCbasedRestraints)
# Separated function for the excluded volume restraint
if use_excluded_volume:
# print "\nEnforcing the excluded_volume_restraints"
# This function is specific for IMP
excluded_volume_kforce = CONFIG[
'ev_kforce'] if 'ev_kforce' in CONFIG else CONFIG['kforce']
evr = add_excluded_volume_restraint(model, model['particles'],
excluded_volume_kforce)
if verbose == 1:
try:
print "Total number of restraints: %i" % (
model['model'].get_number_of_restraints())
if use_HiC:
print len(HiCbasedRestraints)
except:
print "Total number of restraints: %i" % (
model['restraints'].get_number_of_restraints()) # 2.6.1 compat
if use_HiC:
print len(HiCbasedRestraints)
# Separated function for the Conjugate gradient optimization
if verbose == 1:
print(
"\nPerforming the optimization of the Hi-C-based restraints "
"using conjugate gradient optimisation\n")
conjugate_gradient_optimization(model, log_energies)
if verbose >= 1:
if verbose >= 2 or not rand_init % 100:
print 'Model %s IMP Objective Function: %s' % (rand_init,
log_energies[-1])
x, y, z, radius = (FloatKey("x"), FloatKey("y"), FloatKey("z"),
FloatKey("radius"))
result = IMPmodel({
'log_objfun': log_energies,
'objfun': log_energies[-1],
'x': [],
'y': [],
'z': [],
'radius': None,
'cluster': 'Singleton',
'rand_init': str(rand_init)
})
for part in model['particles'].get_particles():
result['x'].append(
part.get_value(x) * float(CONFIG['resolution'] * CONFIG['scale']))
result['y'].append(
part.get_value(y) * float(CONFIG['resolution'] * CONFIG['scale']))
result['z'].append(
part.get_value(z) * float(CONFIG['resolution'] * CONFIG['scale']))
# for part in model['particles'].get_particles():
# result['x'].append(part.get_value(x))
# result['y'].append(part.get_value(y))
# result['z'].append(part.get_value(z))
if verbose == 3:
print(part.get_name(), part.get_value(x), part.get_value(y),
part.get_value(z), part.get_value(radius))
# gets radius from last particle, assuming that all are the same
# include in the loop when radius changes... should be a list then
#result['radius'] = part.get_value(radius)
#result['radius'] = RADIUS
result['radius'] = float(CONFIG['resolution'] * CONFIG['scale']) / 2
return result # rand_init, result