def config(state): # how should we define energies of conformations? confEcalc = osprey.ConfEnergyCalculator( state.confSpace, ecalc, referenceEnergies = osprey.ReferenceEnergies(state.confSpace, ecalc), # use the "AllOnPairs" setting to get tighter energy bounds # tighter energy bounds make SOFEA run much faster! energyPartition = osprey.EnergyPartition.AllOnPairs ) # train forcefield approximations to make energy calculations go much faster amat = osprey.ApproximatorMatrix( confEcalc, # save the approximator matrix to disk so we don't have to calculate it again later cacheFile = 'sofea.%s.amat' % state.name, ) # update the conformation energy calculator with the forcefield approximations confEcalc = osprey.ConfEnergyCalculator( state.confSpace, ecalc, # copy settings from the previous confEcalc referenceEnergies = confEcalc.eref, energyPartition = confEcalc.epart, # give the new confEcalc the approximator matrix amat = amat, # how much error can we tolerate per conformation? approximationErrorBudget = 0.1 # kcal/mol ) # make the SOFEA config return osprey.SOFEA_StateConfig( # calculate the energy matrix emat = osprey.EnergyMatrix( confEcalc, # save the emat to disk so we don't have to calculate it again later cacheFile = 'sofea.%s.emat' % state.name, # correct overly optimistic energy bounds with triples to get even tighter energy bounds! tripleCorrectionThreshold = 0.0 ), confEcalc = confEcalc )
def make_pfunc_factory(conf_space, ffparams, numcores, eps, algo_index,
emat_cache_pattern, data):
"""make_pfunc_factory
Return a PartitionFunctionFactory object for the input confspace, which we can use to make
partition functions for various sequences.
"""
parallelism = osprey.Parallelism(cpuCores=numcores)
data['numCpus'] = numcores
# how should we compute energies of molecules?
minimizingEcalc = osprey.EnergyCalculator(conf_space,
ffparams,
parallelism=parallelism,
isMinimizing=True)
# Compute reference energies
eref = osprey.ReferenceEnergies(conf_space, minimizingEcalc)
#Create a minimizing energy calculator
confEcalcMinimized = osprey.ConfEnergyCalculator(conf_space,
minimizingEcalc,
referenceEnergies=eref)
# we need rigid energies too for many algorithms
rigidEcalc = osprey.SharedEnergyCalculator(minimizingEcalc,
isMinimizing=False)
rigidConfEcalc = osprey.ConfEnergyCalculatorCopy(confEcalcMinimized,
rigidEcalc)
confEcalcRigid = rigidConfEcalc
# Specify the type of partitionFunction
if ALGO_LIST[algo_index] == 'SHARK': # using SHARK*
impt_ecalc = rigidConfEcalc
choose_markstar = False
elif ALGO_LIST[algo_index] == 'MARK': # using MARK*
impt_ecalc = rigidConfEcalc
choose_markstar = True
else: # using Gradient descent pfunc
impt_ecalc = None
choose_markstar = False
pfuncFactory = osprey.PartitionFunctionFactory(
conf_space,
confEcalcMinimized,
emat_cache_pattern,
confUpperBoundcalc=impt_ecalc,
useMARK=choose_markstar)
# Set cache pattern
pfuncFactory.setCachePattern(
'%s/emat.%s.%s' % (XTMP_DIR, emat_cache_pattern, data["design name"]))
print('Cache pattern: %s/emat.%s.%s' %
(XTMP_DIR, emat_cache_pattern, data["design name"]))
return pfuncFactory
def makeo_emat(confspace, ffparams, parallelism, fo, continuous=False):
"""Save energy matrix based on OSPREY objects"""
ecalc = osprey.EnergyCalculator(confspace,
ffparams,
parallelism,
isMinimizing=continuous)
eref = osprey.ReferenceEnergies(confspace, ecalc)
conf_ecalc = osprey.ConfEnergyCalculator(confspace,
ecalc,
referenceEnergies=eref)
emat = osprey.EnergyMatrix(conf_ecalc, cacheFile=fo)
def findGmec(epart):
print("\n\nFinding GMEC with energy partition: %s\n" % epart)
# configure conformation energies with the desired energy partition
confEcalc = osprey.ConfEnergyCalculator(confSpace, ecalc, energyPartition=epart)
# find the GMEC
emat = osprey.EnergyMatrix(confEcalc)
astar = osprey.AStarTraditional(emat, confSpace)
gmec = osprey.GMECFinder(astar, confEcalc).find()
def configure_bbk(instance, minimizingEcalc, type_string, id_obj, algo_index):
"""Configure the energy calculators and info for BBK* instance"""
for info in instance.confSpaceInfos():
# Compute reference energies
eref = osprey.ReferenceEnergies(info.confSpace, minimizingEcalc)
#Create a minimizing energy calculator
info.confEcalcMinimized = osprey.ConfEnergyCalculator(
info.confSpace, minimizingEcalc, referenceEnergies=eref)
# BBK* needs rigid energies too
rigidEcalc = osprey.SharedEnergyCalculator(minimizingEcalc,
isMinimizing=False)
rigidConfEcalc = osprey.ConfEnergyCalculatorCopy(
info.confEcalcMinimized, rigidEcalc)
info.confEcalcRigid = rigidConfEcalc
# Specify the input for the partition functions. Providing the confUpperBoundcalc turns on SHARK*
if ALGO_LIST[algo_index] == 'SHARK':
impt_ecalc = rigidConfEcalc
choose_markstar = False
elif ALGO_LIST[algo_index] == 'MARK':
impt_ecalc = rigidConfEcalc
choose_markstar = True
else:
impt_ecalc = None
choose_markstar = False
info.pfuncFactory = osprey.PartitionFunctionFactory(
info.confSpace,
info.confEcalcMinimized,
info.id,
confUpperBoundcalc=impt_ecalc,
useMARK=choose_markstar)
# Set cache pattern
info.pfuncFactory.setCachePattern('%s/emat.%s.%s' %
(XTMP_DIR, info.id, id_obj))
print('Cache pattern: %s/emat.%s.%s' % (XTMP_DIR, info.id, id_obj))
# compute the energy matrices
info.ematMinimized = info.pfuncFactory.getOrMakeEmat(
info.confEcalcMinimized, 'minimized')
info.ematRigid = info.pfuncFactory.getOrMakeEmat(
info.confEcalcRigid, 'rigid')
# Updating energy matrix?
#info.ematCorrected =\
#osprey.c.ematrix.UpdatingEnergyMatrix(
#info.confSpace,
#info.ematMinimized,
#info.confEcalcMinimized
#)
# how should confs be ordered and searched? (don't forget to capture emat by using a defaulted argument)
def makeAStar_min(rcs, emat=info.ematMinimized):
return osprey.AStarTraditional(emat, rcs, showProgress=True)
info.confSearchFactoryMinimized =\
osprey.KStar.ConfSearchFactory(makeAStar_min)
def makeRigidAStar(rcs, emat=info.ematRigid):
return osprey.AStarTraditional(emat, rcs, showProgress=True)
info.confSearchFactoryRigid =\
osprey.KStar.ConfSearchFactory(makeRigidAStar)
paste = osprey.Paste(
complexConfSpace,
numPDBs=15,
epsilon=
0.68, # you proabably want something more precise in your real designs
useWindowCriterion=True,
writeSequencesToConsole=True,
writeSequencesToFile='paste.results.tsv',
mutFile='mut.txt')
# configure PAStE inputs for the conf space
# how should we define energies of conformations?
eref = osprey.ReferenceEnergies(paste.protein.confSpace, ecalc)
paste.protein.confEcalc = osprey.ConfEnergyCalculator(paste.protein.confSpace,
ecalc,
referenceEnergies=eref)
# compute the energy matrix
emat = osprey.EnergyMatrix(paste.protein.confEcalc,
cacheFile='emat.%s.dat' % paste.protein.id)
# how should confs be ordered and searched? (don't forget to capture emat by using a defaulted argument)
def makeAStar(rcs, emat=emat):
return osprey.AStarTraditional(emat, rcs, showProgress=False)
paste.protein.confSearchFactory = osprey.Paste.ConfSearchFactory(makeAStar)
# run PAStE
strand = osprey.Strand('1CC8.ss.pdb')
strand.flexibility['A2'].setLibraryRotamers('ALA', 'GLY')
strand.flexibility['A3'].setLibraryRotamers(osprey.WILD_TYPE, 'VAL')
strand.flexibility['A4'].setLibraryRotamers(osprey.WILD_TYPE)
# make the conf space
confSpace = osprey.ConfSpace(strand)
# choose a forcefield
ffparams = osprey.ForcefieldParams()
# how should we compute energies of molecules?
ecalc = osprey.EnergyCalculator(confSpace, ffparams)
# how should we define energies of conformations?
confEcalc = osprey.ConfEnergyCalculator(confSpace, ecalc)
# compute the energy matrix
emat = osprey.EnergyMatrix(confEcalc)
# run DEE with just steric pruning
pmat = osprey.DEE(confSpace, emat, showProgress=True)
# or run DEE with Goldstein pruning
#i0 = 10.0 # kcal/mol
#pmat = osprey.DEE(confSpace, emat, showProgress=True, singlesGoldsteinDiffThreshold=i0, pairsGoldsteinDiffThreshold=i0)
# find the best sequence and rotamers
astar = osprey.AStarMPLP(emat, pmat)
gmec = osprey.GMECFinder(astar, confEcalc).find()
def confEcalcFactory(confSpace, ecalc):
eref = osprey.ReferenceEnergies(confSpace, ecalc)
return osprey.ConfEnergyCalculator(confSpace,
ecalc,
referenceEnergies=eref)
bbkstar = osprey.BBKStar(
proteinConfSpace,
ligandConfSpace,
complexConfSpace,
numBestSequences=2,
epsilon=
0.99, # you proabably want something more precise in your real designs
writeSequencesToConsole=True,
writeSequencesToFile='bbkstar.results.tsv')
# configure BBK* inputs for each conf space
for info in bbkstar.confSpaceInfos():
# how should we define energies of conformations?
eref = osprey.ReferenceEnergies(info.confSpace, minimizingEcalc)
info.confEcalcMinimized = osprey.ConfEnergyCalculator(
info.confSpace, minimizingEcalc, referenceEnergies=eref)
# compute the energy matrix
ematMinimized = osprey.EnergyMatrix(info.confEcalcMinimized,
cacheFile='emat.%s.dat' % info.id)
# how should confs be ordered and searched?
# (since we're in a loop, need capture variables above by using defaulted arguments)
def makeAStarMinimized(rcs, emat=ematMinimized):
return osprey.AStarTraditional(emat, rcs, showProgress=False)
info.confSearchFactoryMinimized = osprey.BBKStar.ConfSearchFactory(
makeAStarMinimized)
# BBK* needs rigid energies too
confEcalcRigid = osprey.ConfEnergyCalculatorCopy(info.confEcalcMinimized,
osprey.StateUnmutable('ligand', osprey.ConfSpace(ligand)),
osprey.StateMutable('complex', osprey.ConfSpace([design, ligand]))
])
# train a LUTE model for each state
for state in confSpace.states:
# how should we compute energies of molecules?
ecalc = osprey.EnergyCalculator(state.confSpace,
ffparams,
parallelism=parallelism)
# how should we define energies of conformations?
eref = osprey.ReferenceEnergies(state.confSpace, ecalc)
confEcalc = osprey.ConfEnergyCalculator(state.confSpace,
ecalc,
referenceEnergies=eref)
# calculate the energy matrix so we can do DEE
emat = osprey.EnergyMatrix(
confEcalc,
# save it to disk so we don't have to calculate it again later
cacheFile='sofea.%s.emat' % state.name)
# Good LUTE fits rely heavily on DEE pruning, so use the best available DEE options here.
# With interval pruning, we prune only conformations whose energies are more than X higher
# than the lower bound of the conformation with the minimum lower bound.
# If we sort the whole conformation space by lower bounds, interval pruning is like keeping
# only conformations in the bottom X kcal/mol slice of the sorted list.
pruningInterval = 20.0 # kcal/mol
pmat = osprey.DEE(
# how should we compute energies of molecules?
# (give the complex conf space to the ecalc since it knows about all the templates and degrees of freedom)
numCPUs = 40
parallelism = osprey.Parallelism(cpuCores=numCPUs)
ecalc = osprey.EnergyCalculator(confSpace, ffparams, parallelism=parallelism)
rigidEcalc = osprey.EnergyCalculator(confSpace,
ffparams,
parallelism=parallelism,
isMinimizing=False)
# how should we define energies of conformations?
eref = osprey.ReferenceEnergies(confSpace, ecalc)
erefRigid = osprey.ReferenceEnergies(confSpace, rigidEcalc)
confECalc = osprey.ConfEnergyCalculator(confSpace,
ecalc,
referenceEnergies=eref)
confRigidECalc = osprey.ConfEnergyCalculator(confSpace,
rigidEcalc,
referenceEnergies=eref)
# variables for EWAKStar
epsilon = 0.68
numTopSeqs = 5
maxPFConfs = 500
seqFilterOnly = False
wtBenchmark = False
useMaxMutable = True
#want to change maxMutable for various designs
maxMutable = 1
numFilteredSeqs = 10000
# define a strand
strand = osprey.Strand('1CC8.ss.pdb')
strand.flexibility['A2'].setLibraryRotamers('ALA', 'GLY')
strand.flexibility['A3'].setLibraryRotamers(osprey.WILD_TYPE, 'VAL')
strand.flexibility['A4'].setLibraryRotamers(osprey.WILD_TYPE)
# make the conf space
confSpace = osprey.ConfSpace(strand)
# choose a forcefield
ffparams = osprey.ForcefieldParams()
# compute the reference energies (with res entropy)
ecalc = osprey.EnergyCalculator(confSpace, ffparams)
eref = osprey.ReferenceEnergies(confSpace, ecalc, addResEntropy=True)
# define conf energy, with res entropy
confEcalc = osprey.ConfEnergyCalculator(confSpace, ecalc, referenceEnergies=eref, addResEntropy=True)
# use the reference energies for energy matrix computation
emat = osprey.EnergyMatrix(confEcalc)
# Optional: using edge MPLP when using reference energies can make A* faster
astar = osprey.AStarMPLP(emat, confSpace, updater=osprey.EdgeUpdater())
# find the best sequence and rotamers
gmec = osprey.GMECFinder(astar, confEcalc).find()
