def _minimize_step(sf, pose):
mmap = MoveMap()
mmap.set_bb(True)
mmap.set_chi(False)
mmap.set_jump(True)
min_mover = MinMover(mmap, sf, "lbfgs_armijo_nonmonotone", 0.0001, True)
min_mover.max_iter(1000)
min_mover.apply(pose)
def relax(pose):
sf = create_score_function("ref2015")
sf.set_weight(rosetta.core.scoring.atom_pair_constraint, 5)
sf.set_weight(rosetta.core.scoring.dihedral_constraint, 1)
sf.set_weight(rosetta.core.scoring.angle_constraint, 1)
mmap = MoveMap()
mmap.set_bb(True)
mmap.set_chi(True)
mmap.set_jump(True)
relax = rosetta.protocols.relax.FastRelax()
relax.set_scorefxn(sf)
relax.max_iter(200)
relax.dualspace(True)
relax.set_movemap(mmap)
switch = SwitchResidueTypeSetMover("fa_standard")
switch.apply(pose)
relax.apply(pose)
def __init__(self, net, pose, fix=None, glycinate=False, max_iter=100,
n_moves=1, scorefxn=None, kT=0.1, k=15, dropout=0.5, relax=True):
super(NetPackMover, self).__init__(net, pose, k=k, dropout=dropout)
self.relax = relax
self.n_moves = n_moves
self.max_iter = max_iter
self.kT = kT
self.scorefxn = scorefxn
self.fix = fix if fix is not None else []
self.step = 0
self.fastrelax = ...
if relax or glycinate:
fastrelax = FastRelax()
fastrelax.set_scorefxn(self.scorefxn)
mm = MoveMap()
mm.set_bb(False)
mm.set_chi(True)
#for fixed in self.fix:
# mm.set_chi(fixed, False)
fastrelax.set_movemap(mm)
self.fastrelax = fastrelax
if glycinate:
self.dropout = 1.0
mask = torch.tensor([
1 - int(self.fixed_position(idx))
for idx in range(len(pose.sequence()))
], dtype=torch.bool)
for idx, residue in enumerate(pose.residues):
residue_name = self.sample_residue(pose, idx, mask=mask, argmax=True)
if not self.fixed_position(idx):
mutate_residue(pose, idx + 1, residue_name, pack_radius=10.0, pack_scorefxn=scorefxn)
mask[idx] = 0
self.dropout = dropout
self.fastrelax.apply(pose)
self.monte_carlo = MonteCarlo(pose, scorefxn, kT)
self.glycinate = glycinate
def movemap(pose, PDB_out=False):
"""
Demonstrates the syntax necessary for basic usage of the MoveMap object
performs these changes with a demonstrative backbone minimization
using <pose> and writes structures to PDB files if <PDB_out> is True
"""
#########
# MoveMap
# a MoveMap encodes what data is allowed to change in a Pose, referred to as
# its degrees of freedom
# a MoveMap is separate from a Pose and is usually required by a Mover so
# that the correct degrees of freedom are manipulated, in this way,
# MoveMap and Pose objects often work in parallel
# several MoveMap's can correspond to the same Pose
# a MoveMap stores information on a per-residue basis about the
# backbone ({phi, psi, omega}) and chi ({chi_i}) torsion angle sets
# the MoveMap can only set these sets of torsions to True or False, it
# cannot set freedom for the individual angles (such as phi free and psi
# fixed)
# the MoveMap has no upper-limit on its residue information, it defaults to
# all residues (up to residue 99999999) backbone and chi False
# you can view the MoveMap per-residue torsion settings by using the
# MoveMap.show( Pose.total_residue() ) method (the input argument is the
# highest residue to output, it does not support viewing a range)
pose_move_map = MoveMap()
# change all backbone torsion angles
pose_move_map.set_bb(True)
# change all chi angle torsion angles (False by default)
pose_move_map.set_chi(False)
# change a single backbone torsion angles
#pose_move_map.set_bb(1, True) # example syntax
# change a single residue's chi torsion angles
#pose_move_map.set_chi(1, True) # example syntax
pose_move_map.show(pose.total_residue())
# perform gradient based minimization on the "median" residues, this
# method (MinMover) determines the gradient of an input pose using a
# ScoreFunction for evaluation and a MoveMap to define the degrees of
# freedom
# create a standard ScoreFunction
scorefxn = get_fa_scorefxn(
) # create_score_function_ws_patch('standard', 'score12')
# redefine the MoveMap to include the median half of the residues
# turn "off" all backbone torsion angles
pose_move_map.set_bb(False) # reset to backbone False
# turn "on" a range of residue backbone torsion angles
pose_move_map.set_bb_true_range(int(pose.total_residue() / 4),
int(pose.total_residue() * 3 / 4))
# create the MinMover
minmover = protocols.minimization_packing.MinMover()
minmover.score_function(scorefxn)
minmover.movemap(pose_move_map)
# create a copy of the pose
test_pose = Pose()
test_pose.assign(pose)
# apply minimization
scorefxn(test_pose) # to prevent verbose output on the next line
pymover = PyMOLMover()
#### uncomment the line below and "comment-out" the two lines below to
#### export the structures into different PyMOL states of the same object
#pymover.keep_history = True # enables viewing across states
#### comment-out the line below, changing PDBInfo names tells the
#### PyMOLMover to produce new objects
test_pose.pdb_info().name('original')
pymover.apply(test_pose)
print('\nPre minimization score:', scorefxn(test_pose))
minmover.apply(test_pose)
if PDB_out:
test_pose.dump_pdb('minimized.pdb')
print('Post minimization score:', scorefxn(test_pose))
#### comment-out the line below
test_pose.pdb_info().name('minimized')
pymover.apply(test_pose)
cleaning.cleanATOM(name)
clean_name = pdb_file + ".clean.pdb"
#rna_set = core.chemical.ChemicalManager.get_instance().residue_type_set("coarse_rna").get()
#rna_set = core.chemical.ChemicalManager.get_instance().residue_type_set("fa_standard")
initial_pose = pose_from_pdb(clean_name)
print initial_pose
print toolbox.get_secstruct(initial_pose)
#Set up ScoreFunction
sf = get_fa_scorefxn()
#Set up MoveMap.
mm = MoveMap()
#change these for more or less flexability
mm.set_bb(True)
mm.set_chi(True)
#Pack and minimize initial pose to remove clashes.
pre_pre_packing_score = sf(initial_pose)
print "1"
task = standard_packer_task(initial_pose)
task.restrict_to_repacking()
task.or_include_current(True)
print "2"
pack_rotamers_mover = protocols.simple_moves.RotamerTrialsMover(sf, task)
pack_rotamers_mover.apply(initial_pose)
print "3"
min_mover = protocols.simple_moves.MinMover()
print "4"
pymol = PyMOLMover() # If Pymol server is running, centroid stage will display
loop_begin = 145
loop_end = 155
loop_cut = 150
my_loop = protocols.loops.Loop(loop_begin, loop_end, loop_cut)
my_loops = protocols.loops.Loops()
my_loops.add_loop(my_loop)
print(my_loop)
protocols.loops.set_single_loop_fold_tree(p, my_loop)
movemap = MoveMap()
movemap.set_bb_true_range(loop_begin, loop_end)
movemap.set_chi(True)
print(p.fold_tree())
print("setting up movers")
# use the KinematicMover explicitly in centroid stage
kic_mover = KinematicMover()
#centroid/fullatom conversion movers
to_centroid = protocols.simple_moves.SwitchResidueTypeSetMover('centroid')
to_fullatom = protocols.simple_moves.SwitchResidueTypeSetMover('fa_standard')
recover_sidechains = protocols.simple_moves.ReturnSidechainMover(starting_p)
#set up sidechain packer movers
task_pack = core.pack.task.TaskFactory.create_packer_task(starting_p)
task_pack.restrict_to_repacking()
def sample_single_loop_modeling(pdb_filename,
loop_begin,
loop_end,
loop_cutpoint,
frag_filename,
frag_length,
outer_cycles_low=2,
inner_cycles_low=5,
init_temp_low=2.0,
final_temp_low=0.8,
outer_cycles_high=5,
inner_cycles_high=10,
init_temp_high=2.2,
final_temp_high=0.6,
jobs=1,
job_output='loop_output'):
"""
Performs simple single loop construction on the input <pdb_filename>
with loop from <loop_begin> to <loop_end> with a
cutpoint at <loop_cutpoint> using fragments of length <frag_length>
in the file <frag_filename>. <jobs> trajectories are performed,
each using a low resolution (centroid) simulated annealing with
<outer_cycles> rounds and <inner_cycles> steps per round decrementing
"temperature" from <init_temp> to <final_temp> geometrically.
Output structures are named <job_output>_(job#).pdb.
"""
# 1. create a pose from the desired PDB file
p = Pose()
pose_from_file(p, pdb_filename)
# 2. create a reference copy of the pose in fullatom
starting_p = Pose()
starting_p.assign(p)
#### if you are constructing multiple loops simultaneously, changes will
#### occur in most of the steps below
# 3. create the Loop object
# (note: Loop objects merely specify residues, they contain no
# conformation data)
my_loop = protocols.loops.Loop(loop_begin, loop_end, loop_cutpoint)
#### if using multiple loops, add additional Loop objects
# 4. use the Loop to set the pose FoldTree
protocols.loops.set_single_loop_fold_tree(p, my_loop)
#### alternate FoldTree setup, if you uncomment the lines below,
#### comment-out the set_single_loop_foldtree line above (line 189)
#### -create an empty FoldTree
#ft = FoldTree()
#### -make it a single edge the length of pose
#ft.simple_tree(p.total_residue())
#### -insert a jump corresponding to the single loop region
#ft.add_jump(loop_begin - 2, loop_end + 2, loop_cutpoint)
#### -give the pose this FoldTree (set it to this object), this will
#### erase any previous FoldTree held by the pose
#p.fold_tree(ft)
#### there is also a fold_tree_from_loops method in exposed which sets up
#### a FoldTree but it is different from set_single_loop_foldtree in
#### that is creates jumps +/- 1 residue from their corresponding loop
#### endpoints and requires a third argument, the FoldTree to setup
# 5. sets the cut-point residues as cut-point variants
protocols.loops.add_single_cutpoint_variant(p, my_loop)
# 6. create the MoveMap, allow the loop region backbone and
# all chi torsions to be free
movemap = MoveMap()
movemap.set_bb_true_range(loop_begin, loop_end)
movemap.set_chi(True) # sets all chi torsions free
# 7. setup the fragment Mover
# this "try--except" is used to catch improper fragment files
try:
fragset = core.fragment.ConstantLengthFragSet(frag_length,
frag_filename)
#### the ConstantLengthFragSet is overloaded, this same
#### ConstantLengthFragSet can be obtained with different syntax
# to obtain custom fragments, see Generating Fragment Files below
except:
raise IOError('Make sure frag_length matches the fragments in\n\
frag_file and that frag_file is valid')
fragment_mover = protocols.simple_moves.ClassicFragmentMover(
fragset, movemap)
# 8. create a Mover for loop modeling using CCD (low resolution)
ccd_closure = protocols.loops.loop_closure.ccd.CCDLoopClosureMover(
my_loop, movemap)
# 9. create ScoreFunctions
# for centroid, use the default centroid ScoreFunction with chainbreak on
scorefxn_low = create_score_function('cen_std')
# the chainbreak ScoreType exists to penalize broken bonds
# try creating a broken pose in the interpreter and use a ScoreFunction
# with a chainbreak score to investigate its impact, the score is 0.0
# except when a bond is broken
# this penalizes failures caused by CCD failing to close the loop
scorefxn_low.set_weight(core.scoring.chainbreak, 1)
# for fullatom, used for packing and scoring final output
scorefxn_high = get_fa_scorefxn(
) # create_score_function_ws_patch('standard', 'score12')
# 10. setup sidechain packing Mover
task_pack = core.pack.task.TaskFactory.create_packer_task(starting_p)
task_pack.restrict_to_repacking() # prevents design, packing only
task_pack.or_include_current(True) # considers original sidechains
pack = protocols.minimization_packing.PackRotamersMover(
scorefxn_high, task_pack)
# 11. setup the high resolution refinement
# by creating a Loops object,
# (note: Loops is basically a list of Loop objects),
sample_loops = protocols.loops.Loops()
# giving it the loop to remodel,
sample_loops.add_loop(my_loop)
# and creating a fullatom CCD Mover (high resolution)
# this Mover is somewhat abnormal since it handles everything itself, it:
# -creates its own MoveMap for the loop regions
# -creates its own ScoreFunction (default to get_fa_scorefxn())
# -creates its own FoldTree for the pose based on the loops
# -creates its own MonteCarlo object for monitoring the pose
# -performs "simulated annealing" with 3 outer cycles and 90 inner
# cycles, very similar to the protocol outlined ere
# -creates its own backbone Movers (SmallMover, ShearMover)
# -creates its own PackRotamersMover, it does NOT restrict repacking
# to the loop regions and can alter all sidechain conformations
loop_refine = LoopMover_Refine_CCD(sample_loops)
# some of these parameters or objects can be set but the protocol
# executed by this Mover is effectively untouchable
#loop_refine.set_score_function(scorefxn_high) # in beta v2 and above
loop_refine.temp_initial(init_temp_high)
loop_refine.temp_final(init_temp_high)
loop_refine.outer_cycles(outer_cycles_high)
loop_refine.max_inner_cycles(inner_cycles_high)
# 12. create centroid <--> fullatom conversion Movers
to_centroid = SwitchResidueTypeSetMover('centroid')
to_fullatom = SwitchResidueTypeSetMover('fa_standard')
# and a Mover to recover sidechain conformations
# when a protocol samples backbone torsion space in centroid,
# the sidechain conformations are neglected, when it is transferred
# to fullatom, we typically set the sidechain conformations to their
# "original" values and perform sidechain packing,
# a ReturnSidechainMover saves a pose's sidechains (in this case
# staring_pose) and when applied, inserts these conformations
# into the input pose
recover_sidechains = protocols.simple_moves.ReturnSidechainMover(
starting_p)
# 13. create a reference copy of the pose in centroid
# the first stage of each trajectory is in centroid
# so a centroid reference is needed and the pose must start in centroid
to_centroid.apply(p)
starting_p_centroid = Pose()
starting_p_centroid.assign(p)
# 14. create the geometric "temperature" increment for simulated annealing
gamma = pow((final_temp_low / init_temp_low),
(1.0 / (outer_cycles_low * inner_cycles_low)))
# 15. create a PyMOLMover for exporting structures to PyMOL
pymov = PyMOLMover()
# uncomment the line below to load structures into successive states
#pymov.keep_history(True)
scorefxn_high(starting_p) # for exporting the scores
pymov.apply(starting_p)
pymov.send_energy(starting_p)
# 16. create a (Py)JobDistributor
# a PyJobDistributor uses the job_output argument to name all output files
# and performs the specified number (int) of jobs
# a ScoreFunction is required since the PyJobDistributor output .fasc file
# contains scoring information about each output PDB
jd = PyJobDistributor(job_output, jobs, scorefxn_high)
jd.native_pose = starting_p
# 17. perform the loop modeling protocol
counter = 0 # for exporting to PyMOL
while not jd.job_complete:
# a. set necessary variables for the new trajectory
# -reload the starting pose (centroid)
p.assign(starting_p_centroid)
# -change the pose's PDBInfo.name, for exporting to PyMOL
counter += 1
p.pdb_info().name(job_output + '_' + str(counter) + '_cen')
# -reset the starting "temperature" (to init_temp)
kT = init_temp_low
# -create a MonteCarlo object for this trajectory
# a MonteCarlo object assesses pass/fail by the Metropolis Criteria
# and also records information on the lowest scoring pose
mc = MonteCarlo(p, scorefxn_low, kT)
# b. "randomize" the loop
#### this section may change if you intend to use multiple loops or
#### alter the sampling method to "randomize" the loop
# -by breaking it open,
for i in range(loop_begin, loop_end + 1):
p.set_phi(i, -180)
p.set_psi(i, 180)
pymov.apply(p)
# -and then inserting fragments
# the number of insertions performed is somewhat arbitrary
for i in range(loop_begin, loop_end + 1):
fragment_mover.apply(p)
pymov.apply(p)
####
# low resolution loop modeling:
# c. simulated annealing incrementing kT geometrically
# from init_temp to final_temp
#### this section may change if you intend to use multiple loops or
#### alter the sampling method for low resolution modeling
for i in range(1, outer_cycles_low + 1):
# -start with the lowest scoring pose
mc.recover_low(p) # loads mc's lowest scoring pose into p
# -take several steps of in the simulated annealing by
for j in range(1, inner_cycles_low + 1):
# >increasing the "temperature"
kT = kT * gamma
mc.set_temperature(kT)
# >inserting a fragment,
fragment_mover.apply(p)
pymov.apply(p)
# >performing CCD,
ccd_closure.apply(p)
pymov.apply(p)
# >and assessing the Metropolis Criteria
mc.boltzmann(p)
####
# the LoopMover_Refine_CCD makes A LOT of moves, DO NOT expect to
# see useful results if you use the PyMOLMover keep_history option, the large
# number of intermediates will slow processing to a halt
# d. convert the best structure (lowest scoring) into fullatom by:
# -recovering the best (centroid) structure (lowest scoring),
mc.recover_low(p) # loads mc's lowest scoring pose into p
# -switching the ResidueTypeSet to fullatom (from centroid),
to_fullatom.apply(p)
# -recovering the original sidechain conformations,
recover_sidechains.apply(p)
# -and packing the result (since the backbone conformation has changed)
pack.apply(p)
pymov.apply(p)
p.pdb_info().name(job_output + '_' + str(counter) + '_fa')
# high-resolution refinement:
#### this section may change if you intend to use multiple loops or
#### alter the sampling method for high resolution refinement
# e. apply the LoopMover_Refine_CCD
loop_refine.apply(p)
# f. output the decoy (pose result from this trajectory)
# include the loop RMSD (Lrsmd)
# -output a PDB file using the PyJobDistributor
lrms = protocols.loops.loop_rmsd(p, starting_p, sample_loops, True)
jd.additional_decoy_info = ' Lrmsd: ' + str(lrms)
jd.output_decoy(p)
# -export the structure to PyMOL
pymov.apply(p)
pymov.send_energy(p)
def mutate(pdb_name, n_muts=100):
OUT = open(mydir + '/data/deltas.txt', 'w')
#takes name of pdb file without the extention
init(extra_options='-mute basic -mute core')
# Constants
PACK_RADIUS = 10.0
#Amino acids, notice there is no C
AAs = ("A", "D", "E", "F", "G", "H", "I", "K", "L", "M", "N", "P", "Q",
"R", "S", "T", "V", "W", "Y")
#Number of mutations to accept
max_accept_mut = 600
#Population size
N = 100
#Beta (temp term)
beta = 1
pdb = pdb_name + ".pdb"
cleanATOM(pdb)
pdb_clean_name = pdb_name + '.clean.pdb'
initial_pose = pose_from_pdb(pdb_clean_name)
#Set up ScoreFunction
sf = get_fa_scorefxn()
#Set up MoveMap.
mm = MoveMap()
#change these for more or less flexability
mm.set_bb(True)
mm.set_chi(True)
#Pack and minimize initial pose to remove clashes.
pre_pre_packing_score = sf(initial_pose)
task = standard_packer_task(initial_pose)
task.restrict_to_repacking()
task.or_include_current(True)
pack_rotamers_mover = RotamerTrialsMover(sf, task)
pack_rotamers_mover.apply(initial_pose)
min_mover = MinMover()
min_mover.movemap(mm)
min_mover.score_function(sf)
min_mover.min_type('dfpmin_armijo_nonmonotone')
min_mover.apply(initial_pose)
post_pre_packing_score = sf(initial_pose)
#Set threshold for selection
threshold = pre_pre_packing_score / 2
#number of residues to select from
n_res = initial_pose.total_residue()
max_accept_mut = 1500
#start sim
i = 0
gen = 0
columns = ['Number', 'Selection', 'P_fix', 'P_fix_MH']
print >> OUT, '\t'.join(columns)
while i < max_accept_mut:
#update the number of generations that have pased
gen += 1
#pick a place to mutate
mut_location = random.randint(1, n_res)
#get the amino acid at that position
res = initial_pose.residue(mut_location)
#don't mess with C, just choose again
if res.name1() == 'C':
mut_location = random.randint(1, n_res)
#get the amino acid at that position
res = initial_pose.residue(mut_location)
#choose the amino acid to mutate to
new_mut_key = random.randint(0, len(AAs) - 1)
proposed_res = AAs[new_mut_key]
#don't bother mutating to the same amino acid it just takes more time
if proposed_res == res.name1():
new_mut_key = random.randint(0, len(AAs) - 1)
proposed_res = AAs[new_mut_key]
#make the mutation
mutant_pose = mutate_residue(initial_pose, mut_location, proposed_res,
PACK_RADIUS, sf)
#score mutant
variant_score = sf(mutant_pose)
probability = calc_prob_fix(variant_score, post_pre_packing_score, N,
beta, threshold)
probability_mh = calc_prob_mh(variant_score, post_pre_packing_score, N,
beta, threshold)
f_i = calc_x_fix(post_pre_packing_score, beta, threshold)
f_j = calc_x_fix(variant_score, beta, threshold)
s = math.log(f_j) - math.log(f_i)
#test to see if mutation is accepted
if np.isnan(probability) == True:
continue
rndm = random.random()
if (i < 100):
if (rndm < probability):
initial_pose = mutant_pose
post_pre_packing_score = variant_score
else:
continue
# Assuming 100 burn in phase, take this if out if you want to store everything
else:
data = [str(i), str(s), str(probability), str(probability_mh)]
print >> OUT, '\t'.join(data)
#save name and energy change
#data.append(variant_name + "," + str(variant_score) + "," + str(variant_score - post_pre_packing_score) + "," + str(probability) + "," + str(gen) + "\n")
#pdb_name=str(i)+".pdb"
#mutant_pose.dump_pdb(pdb_name)
print i, s, probability, rndm
#update number of accepts
i += 1
OUT.close()