def parse_arguments():
import argparse
parser = argparse.ArgumentParser(
description='Simulate a SELEX experiment.')
parser.add_argument( '--energy-model', type=file,
help='An energy model to simulate from.')
parser.add_argument( '--pwm', type=file,
help='A pwm to simulate from.')
parser.add_argument( '--sim-sizes', type=int, nargs='+',
help='Number of reads to simulate for each round.')
parser.add_argument( '--prot-conc', type=float, default=7.75e-10,
help='The protein concentration.')
parser.add_argument( '--dna-conc', type=float, default=2e-8,
help='The DNA concentration.')
parser.add_argument( '--random-seed', type=int,
help='Set the random number generator seed.')
parser.add_argument( '--random-seq-pool-size', type=float, default=1e5,
help='The random pool size for the bootstrap.')
parser.add_argument( '--verbose', default=False, action='store_true',
help='Print extra status information.')
args = parser.parse_args()
pyTFbindtools.VERBOSE = args.verbose
if args.random_seed != None:
np.random.seed(args.random_seed)
if args.pwm != None:
pyTFbindtools.log("Loading PWM starting location", 'VERBOSE')
motifs = load_motifs(args.pwm.name)
assert len(motifs) == 1, "Motif file contains multiple motifs"
motif = motifs.values()[0][0]
args.pwm.close()
else:
assert args.energy_model != None, \
"Either --energy-model or --pwm must be specified"
pyTFbindtools.log("Loading energy data", 'VERBOSE')
motif = load_energy_data(args.energy_model.name)
args.starting_energy_model.close()
return ( motif, args.prot_conc, args.dna_conc,
args.sim_sizes,
int(args.random_seq_pool_size) )
def find_best_shift(rnds_and_seqs, ddg_array, ref_energy):
pwm = find_pwm_from_starting_alignment(
rnds_and_seqs[-1], build_pwm_from_energies(ddg_array, ref_energy, -12))
left_shift_pwm = find_pwm_from_starting_alignment(
rnds_and_seqs[-1], np.hstack((np.zeros((4,1)), pwm.T)))
right_shift_pwm = find_pwm_from_starting_alignment(
rnds_and_seqs[-1], np.hstack((pwm.T, np.zeros((4,1)))))
# calculate the entropies, and shift int he direction that gives the
# smallest entropy
left_shift_score = -(
left_shift_pwm[0,:]*np.log(left_shift_pwm[0,:])).sum()
right_shift_score = -(
right_shift_pwm[-1,:]*np.log(right_shift_pwm[-1,:])).sum()
pyTFbindtools.log("Left shift entropy: %.2e" % left_shift_score, 'VERBOSE')
pyTFbindtools.log("Right shift entropy: %.2e" % right_shift_score,'VERBOSE')
if left_shift_score < right_shift_score:
return "LEFT"
else:
return "RIGHT"
def simulate_reads( motif, seq_len, sim_sizes,
dna_conc, prot_conc,
fname_prefix="test",
pool_size = 100000):
ref_energy, ddg_array = motif.build_ddg_array()
chem_pots = est_chem_potentials(
ddg_array, ref_energy, dna_conc, prot_conc,
2*(seq_len-len(motif)+1), len(sim_sizes))
current_pool = np.array([np.random.randint(4, size=seq_len)
for i in xrange(pool_size)])
rnds_and_seqs = []
for rnd, (sim_size, chem_pot) in enumerate(
zip(sim_sizes, chem_pots), start=1):
occs = np.array([motif.est_occ(chem_pot, seq)
for seq in current_pool])
#print current_pool
seq_indices = np.random.choice(
len(current_pool), size=sim_size,
p=occs/occs.sum(), replace=True)
seqs = current_pool[np.array(seq_indices, dtype=int)]
seq_occs = occs[np.array(seq_indices, dtype=int)]
with open("%s_rnd_%i.txt" % (fname_prefix, rnd), "w") as ofp:
for seq in seqs:
print >> ofp, "".join('ACGT'[x] for x in seq)
current_pool = seqs[np.random.choice(
len(seqs), size=pool_size,
p=seq_occs/seq_occs.sum(), replace=True)]
pyTFbindtools.log(
"Finished simulations for round %i" % rnd, level='VERBOSE')
pyTFbindtools.log("Finished Simulations", level='VERBOSE')
pyTFbindtools.log("Ref Energy: %.2f" % ref_energy, level='VERBOSE')
pyTFbindtools.log("Chem Pots: %s" % chem_pots, level='VERBOSE')
pyTFbindtools.log(str(ddg_array), level='VERBOSE')
return
def fit_model(rnds_and_seqs, ddg_array, ref_energy):
opt_path = []
prev_lhd = None
for rnd_num in xrange(min(20,
len(rnds_and_seqs[0][0])-ddg_array.motif_len+1)):
bs_len = ddg_array.motif_len
pyTFbindtools.log("Coding sequences", 'VERBOSE')
partitioned_and_coded_rnds_and_seqs = PartitionedAndCodedSeqs(
rnds_and_seqs, bs_len)
pyTFbindtools.log("Estimating energy model", 'VERBOSE')
( ddg_array, ref_energy, chem_pots, lhd_path, lhd_hat
) = estimate_dg_matrix_with_adadelta(
partitioned_and_coded_rnds_and_seqs,
ddg_array, ref_energy,
dna_conc, prot_conc)
opt_path.append([bs_len, lhd_hat, ddg_array, ref_energy])
pyTFbindtools.log(ddg_array.consensus_seq(), 'VERBOSE')
pyTFbindtools.log("Ref: %s" % ref_energy, 'VERBOSE')
pyTFbindtools.log(
"Mean: %s" % (ref_energy + ddg_array.sum()/3), 'VERBOSE')
pyTFbindtools.log(
"Min: %s" % ddg_array.calc_min_energy(ref_energy), 'VERBOSE')
pyTFbindtools.log(
str(ddg_array.calc_base_contributions().round(2)), 'VERBOSE')
## Old stop criterion
#if prev_lhd != None and prev_lhd + 10 > lhd_hat:
pyTFbindtools.log("Prev: %.2f\tCurr: %.2f\tDiff: %.2f" % (
lhd_path[0], lhd_path[-1], lhd_path[0]-lhd_path[-1]), 'VERBOSE')
if lhd_path[0] + 10 > lhd_path[-1]:
pyTFbindtools.log("Model has finished fitting", 'VERBOSE')
break
# update hte previous likelihood
#prev_lhd = lhd_hat
pyTFbindtools.log("Finding best shift", 'VERBOSE')
shift_type = find_best_shift(rnds_and_seqs, ddg_array, ref_energy)
if shift_type == 'LEFT':
pyTFbindtools.log("Adding left base to motif", level='VERBOSE' )
ddg_array = np.insert(ddg_array, 0, np.zeros(3, dtype='float32')
).view(DeltaDeltaGArray)
elif shift_type == 'RIGHT':
pyTFbindtools.log("Adding right base to motif", level='VERBOSE' )
ddg_array = np.append(ddg_array, np.zeros(3, dtype='float32')).view(
DeltaDeltaGArray)
else:
assert False, "Unrecognized shift type '%s'" % shift_type
ref_energy = ref_energy
for entry in opt_path:
print entry
return ddg_array, ref_energy
def parse_arguments():
import argparse
parser = argparse.ArgumentParser(
description='Estimate energy models from a SELEX experiment.')
parser.add_argument( '--selex-files', nargs='+', type=file, required=True,
help='Files containing SELEX reads.')
parser.add_argument( '--background-sequence', type=file,
help='File containing reads sequenced from round 0.')
parser.add_argument( '--starting-pwm', type=file,
help='A PWM to start from.')
parser.add_argument( '--starting-energy-model', type=file,
help='An energy model to start from.')
parser.add_argument( '--initial-binding-site-len', type=int, default=6,
help='The starting length of the binding site (this will grow)')
parser.add_argument( '--lhd-convergence-eps', type=float, default=1e-8,
help='Convergence tolerance for lhd change.')
parser.add_argument( '--max-iter', type=float, default=1e5,
help='Maximum number of optimization iterations.')
parser.add_argument( '--momentum', type=float, default=0.1,
help='Optimization tuning param (between 0 and 1).')
parser.add_argument( '--random-seed', type=int,
help='Set the random number generator seed.')
parser.add_argument( '--random-seq-pool-size', type=float, default=1e6,
help='The random pool size for the bootstrap.')
parser.add_argument( '--verbose', default=False, action='store_true',
help='Print extra status information.')
parser.add_argument( '--debug-verbose', default=False, action='store_true',
help='Print debug information.')
args = parser.parse_args()
assert not (args.starting_pwm and args.starting_energy_model), \
"Can not set both --starting-pwm and --starting-energy_model"
pyTFbindtools.VERBOSE = args.verbose or args.debug_verbose
pyTFbindtools.DEBUG = args.debug_verbose
pyTFbindtools.selex.CONVERGENCE_MAX_LHD_CHANGE = args.lhd_convergence_eps
pyTFbindtools.selex.MAX_NUM_ITER = int(args.max_iter)
assert args.momentum < 1 and args.momentum >= 0
pyTFbindtools.selex.MOMENTUM = args.momentum
if args.random_seed != None:
np.random.seed(args.random_seed)
pyTFbindtools.log("Loading sequences", 'VERBOSE')
rnds_and_seqs = load_sequences(x.name for x in args.selex_files)
if args.starting_pwm != None:
pyTFbindtools.log("Loading PWM starting location", 'VERBOSE')
motifs = load_motifs(args.starting_pwm)
assert len(motifs) == 1, "Motif file contains multiple motifs"
motif = motifs.values()[0]
args.starting_pwm.close()
elif args.starting_energy_model != None:
pyTFbindtools.log("Loading energy data", 'VERBOSE')
motif = load_energy_data(args.starting_energy_model.name)
args.starting_energy_model.close()
else:
pyTFbindtools.log(
"Initializing starting location from %imer search" % args.initial_binding_site_len,
'VERBOSE')
factor_name = 'TEST'
bs_len = args.initial_binding_site_len
pwm = find_pwm(rnds_and_seqs, args.initial_binding_site_len)
motif = Motif("aligned_%imer" % args.initial_binding_site_len,
factor_name, pwm)
return motif, rnds_and_seqs, int(args.random_seq_pool_size)