def merge_sdflist(lines, datadir=""):
# Get first line
sdffile = datadir + lines[0] + ".sdf"
molobj, atoms, keep_energies, keep_coordinates = worker.get_sdfcontent(
sdffile, rtn_atoms=True)
keep_representations = [
sim.get_representation(atoms, coordinates)
for coordinates in keep_coordinates
]
for line in tqdm(lines[1:]):
# for line in lines[1:]:
sdffile = datadir + line + ".sdf"
from_energies, from_coordinates = worker.get_sdfcontent(sdffile)
from_representations = [
sim.get_representation(atoms, coordinates)
for coordinates in from_coordinates
]
# print("merge", len(from_energies))
# Asymmetrically add new conformers
# idxs_ref = merge_asymmetric_fchl18(atoms,
# from_energies,
# keep_energies,
# from_coordinates,
# keep_coordinates, debug=False)
# find
# energies = np.round(keep_energies, 1)
# idx, = np.where(energies == 14.8)
# print(from_coordinates[idx[0]])
#
# print(idx)
idxs = merge_asymmetric(atoms, from_energies, keep_energies,
from_representations, keep_representations)
for i, idx in enumerate(idxs):
# if conformation already exists, continue
if len(idx) > 0: continue
# Add new unique conformation to collection
keep_energies.append(from_energies[i])
keep_coordinates.append(from_coordinates[i])
keep_representations.append(from_representations[i])
sdfstr = ""
for coordinates in keep_coordinates:
sdfstr += cheminfo.save_molobj(molobj, coordinates)
return sdfstr
def merge_sdflist(lines, datadir=""):
# Get first line
sdffile = datadir + lines[0] + ".sdf"
molobj, atoms, keep_energies, keep_coordinates = worker.get_sdfcontent(
sdffile, rtn_atoms=True)
keep_representations = [
workkernel.FchlRepresentation(atoms, coordinates, **DPARAMETERS)
for coordinates in keep_coordinates
]
for line in tqdm(lines[1:]):
# for line in lines[1:]:
sdffile = datadir + line + ".sdf"
from_energies, from_coordinates = worker.get_sdfcontent(sdffile)
from_representations = [
workkernel.FchlRepresentation(atoms, coordinates, **DPARAMETERS)
for coordinates in from_coordinates
]
# print("merge", len(from_energies))
# Asymmetrically add new conformers
# idxs = merge_asymmetric(atoms,
# from_energies,
# keep_energies,
# from_coordinates,
# keep_coordinates, debug=False)
idxs = merge_asymmetric_objs(from_energies, keep_energies,
from_representations,
keep_representations, **DPARAMETERS)
for i, idx in enumerate(idxs):
# if conformation already exists, continue
if len(idx) > 0: continue
# Add new unique conformation to collection
keep_energies.append(from_energies[i])
keep_coordinates.append(from_coordinates[i])
keep_representations.append(from_representations[i])
sdfstr = ""
for coordinates in keep_coordinates:
sdfstr += cheminfo.save_molobj(molobj, coordinates)
return sdfstr
def run_jobline(origins, molobjs, tordbs, line,
prefix=None,
debug=False,
dump=False):
sep = ","
# TODO multiple molobjs
line = line.strip()
# Locate molobj
line_s = line.split(sep)
molid = int(line_s[0])
molobj = molobjs[molid]
tordb = tordbs[molid]
# deep copy
molobj = copy.deepcopy(molobj)
cheminfo.molobj_set_coordinates(molobj, origins[molid])
if dump:
if prefix is None:
prefix = line.replace(" ", "_").replace(",", ".")
filename = "_tmp_data/{:}.sdf".format(prefix)
# if os.path.exists(filename):
# return [],[]
job_start = time.time()
job_energies, job_coordinates = run_job(molobj, tordb, line)
job_end = time.time()
if debug:
print(line, "-", len(job_energies), "{:5.2f}".format(job_end-job_start), filename)
if dump:
if debug: print("saving {:} confs to".format(len(job_energies)), filename)
fsdf = open(filename, 'w')
for energy, coordinates in zip(job_energies, job_coordinates):
sdfstr = cheminfo.save_molobj(molobj, coordinates)
fsdf.write(sdfstr)
return job_energies, job_coordinates
def calculate_forcefield(molobj, conformer, torsions, origin_angles, delta_angles,
ffprop=None,
ff=None,
delta=10**-7,
coord_decimals=6,
grad_threshold=100):
"""
Disclaimer: lots of hacks, sorry. Let me know if you have an alternative.
Note: There is a artificat where if delta < 10**-16 the FF will find a
*extremely* local minima with very high energy (un-physical)the FF will
find a *extremely* local minima with very high energy (un-physical).
Setting delta to 10**-6 (numerical noise) should fix this.
Note: rdkit forcefield restrained optimization will optimized to a *very*
local and very unphysical minima which the global optimizer cannot get out
from. Truncating the digits of the coordinates to six is a crude but
effective way to slight move the the molecule out of this in a reproducable
way.
"""
if ffprop is None or ff is None:
ffprop, ff = get_forcefield(molobj)
sdfstr = cheminfo.molobj_to_sdfstr(molobj)
molobj_prime, status = cheminfo.sdfstr_to_molobj(sdfstr)
conformer_prime = molobj_prime.GetConformer()
# Setup constrained forcefield
# ffprop_prime, ffc = get_forcefield(molobj_prime)
ffc = ChemicalForceFields.MMFFGetMoleculeForceField(molobj_prime, ffprop)
# Set angles and constrains for all torsions
for i, angle in enumerate(delta_angles):
set_angle = origin_angles[i] + angle
# Set clockwork angle
try: Chem.rdMolTransforms.SetDihedralDeg(conformer_prime, *torsions[i], set_angle)
except: pass
# Set forcefield constrain
ffc.MMFFAddTorsionConstraint(*torsions[i], False,
set_angle-delta, set_angle+delta, 1.0e10)
# minimize constrains
status = run_forcefield(ffc, 500)
# Set result
coordinates = conformer_prime.GetPositions()
coordinates = np.round(coordinates, coord_decimals) # rdkit hack, read description
cheminfo.conformer_set_coordinates(conformer, coordinates)
# minimize global
status = run_forcefield_prime(ff, 700, force=1e-4)
# Get current energy
energy = ff.CalcEnergy()
if status == 0:
grad = ff.CalcGrad()
grad = np.array(grad)
grad_norm = linalg.norm(grad)
if grad_norm > grad_threshold:
status = 4
debug = False
if energy > 1000 and debug:
print(torsions, origin_angles, delta_angles)
print(energy, status)
print("id")
print(id(molobj_prime))
print(id(molobj))
molobj_test, status = cheminfo.sdfstr_to_molobj(sdfstr)
coordinates = conformer.GetPositions()
cheminfo.molobj_set_coordinates(molobj_test, coordinates)
ffprop_t, ff_t = get_forcefield(molobj)
run_forcefield(ff_t, 500)
print(coordinates)
for idxs in torsions:
angle = Chem.rdMolTransforms.GetDihedralDeg(conformer, *idxs)
print("ANGLE 1", angle)
f = open("_test_dumpsdf.sdf", 'w')
sdf = cheminfo.save_molobj(molobj)
f.write(sdf)
# prop, ff = get_forcefield(molobj)
# status = run_forcefield(ff, 500)
conformer = molobj_test.GetConformer()
for idxs in torsions:
angle = Chem.rdMolTransforms.GetDihedralDeg(conformer, *idxs)
print("ANGLE 2",angle)
print(energy, status)
sdf = cheminfo.save_molobj(molobj_test)
f.write(sdf)
f.close()
quit()
# Get current positions
pos = conformer.GetPositions()
return energy, pos, status
def merge_asymmetric(atoms,
energies_x,
energies_y,
coordinates_x,
coordinates_y,
decimals=1,
threshold=0.98,
molobj=None,
debug=False):
"""
"""
coordinates_x = np.asarray(coordinates_x)
coordinates_y = np.asarray(coordinates_y)
energies_x = np.round(energies_x, decimals=decimals)
energies_y = np.round(energies_y, decimals=decimals)
new_unique_energies = np.unique(energies_x)
# Return index from x, with idx of same similarity
# if empty, it is new
rtnidx = [[] for x in range(len(energies_x))]
for uenergy in new_unique_energies:
idx_x, = np.where(energies_x == uenergy)
idx_y, = np.where(energies_y == uenergy)
if debug:
print("amerge", uenergy, len(idx_x), len(idx_y))
if len(idx_y) == 0:
# all unique, continue
continue
if len(idx_y) > 5 and molobj is not None:
atoms_str = [cheminfo.convert_atom(atom) for atom in atoms]
# f = open("_tmp_test.xyz", 'w')
f = open("_tmp_test.sdf", 'w')
coordinates_dump = align_coordinates(coordinates_y[idx_y])
energies = energies_y[idx_y]
for energy, coordinates in zip(energies, coordinates_dump):
# xyzstr = rmsd.set_coordinates(atoms_str, coordinates, title=str(energy))
# xyzstr += "\n"
sdfstr = cheminfo.save_molobj(molobj, coordinates)
f.write(sdfstr)
f.close()
# list of list of idx
similar = merge_asymmetric_similarity(atoms,
coordinates_x[idx_x],
coordinates_y[idx_y],
threshold=threshold)
# convert local similarity to idx_y
for i, sidx in enumerate(similar):
sidx = [idx_y[j] for j in sidx]
sidx = np.asarray(sidx)
similar[i] = sidx
# Create rtn idx
for i, idx in enumerate(idx_x):
rtnidx[idx] = similar[i]
return rtnidx