# -3.63193668, 4.39840442, -1.52225387, -5.26723347, 1.34990893,
# -5.72045913, 1.37062561])
energies = np.array([
-4.00896773, -7.54098962, -0.23402838, 3.14820036, -5.44986465,
-6.16460973, 2.81354172, -3.84467562, -6.34558658, 0.13278807, -4.91226052,
2.78108339
])
for ii in range(100):
SEQUENCE = sequences[ii]
print ii
start = SEQUENCE.index(ms2_hairpin)
end = start + len(ms2_hairpin)
tf_bpp = np.array(
tf.RNA(SEQUENCE, False, list(energies), True, False).get_bpp_full())
vienna_bpp = np.loadtxt(
"vienna_bpp_data/{}_bpp_matrix.txt".format(SEQUENCE))
tf_bpp_matrix = pd.DataFrame(tf_bpp + tf_bpp.transpose(),
index=list(SEQUENCE),
columns=list(SEQUENCE))
vienna_bpp_matrix = pd.DataFrame(vienna_bpp + vienna_bpp.transpose(),
index=list(SEQUENCE),
columns=list(SEQUENCE))
subtitle_fs = 12.
fig, ax = plt.subplots(figsize=(15, 5), ncols=2, nrows=1)
if SEQUENCE in trouble:
import numpy as np
import random
L = 50
sequence_length = np.arange(5, L + 1)
seq_len_3 = []
bpp = []
bpp_grad = []
part = []
grad = []
num = 100
energy = [5.69, 6., 4.09, -7.09]
for l in range(5, L + 1):
seq_len_3.append(7.6e-5 * l**2 + 0.01)
start = time.clock()
for i in range(num):
sequence = ''.join(random.choice('AUGC') for _ in range(l))
parts = tf.RNA(sequence, False, energy, False, False)
end = time.clock()
bpp.append(1e3 * (end - start) / num)
np.savetxt("n3_partition_time.txt", bpp, delimiter=',')
plt.plot(sequence_length, seq_len_3, 'k')
plt.plot(sequence_length, bpp, 'b', linewidth=2)
plt.show()
])
'''BPP TRAINING DATA'''
num_training_examples = 100
actual_bpp = [
] # storing log(bpp) for closing stem of hairpin for each input sequence
sequences = Sequence[:
num_training_examples] #fil.Sequence[:num_training_examples]
for i in range(num_training_examples):
#ms2_prob = 1.
#bp = ms2_hairpin_basepairs[i]
#for mm in range(7):
# ms2_prob *= tf.RNA(sequences[i], False, list(energies), True, False).get_bpp(bp[2*mm], bp[2*mm + 1])
bp = closing_bp_indices[i] #fil.closing_bp_indices[i]
rna.append(
np.log(10.) - tf.RNA(sequences[i], False, list(energies), True,
False).get_log_bpp(bp[0], bp[1]))
#rna.append(10./tf.RNA(sequences[i], False, list(energies), True, False).get_bpp(bp[0], bp[1]))
#rna.append(np.log(10.) - np.log(ms2_prob))
actual_bpp.append(np.log(KDnoligand[i]))
#actual_bpp.append(KDnoligand[i])
print 'finished gathering training data'
rna = np.array(rna)
actual_bpp = np.array(actual_bpp)
RMSD = np.mean((rna - actual_bpp)**2)
plt.title('R101 Synthetic w/ imperfect prior, hella far initial start')
plt.text(4, 17.5, "RMSD = %.2f" % RMSD)
#plt.scatter(actual_bpp, rna2, c='m', label = 'Full')
#plt.scatter(actual_bpp, rna1, c='c', label = 'Kd < 25')
import filtering as fil
rna = []
alpha = 1.
w = 1e-2
energies = np.arange(-10, 2)
'''BPP TRAINING DATA'''
num_training_examples = 1000
actual_bpp = [
] # storing bpp for closing stem of hairpin for each input sequence
#energy_param = p.energies
sequences = fil.Sequence[:num_training_examples] #p.training_sequences[:10]
for i in range(num_training_examples):
rna.append(tf.RNA(sequences[i], False, list(energies), True, False))
actual_bpp.append(np.log(1e-9) - np.log(fil.KDnoligand[i]))
print 'finished gathering training data'
guess = np.zeros(12)
def cost(param, i, j):
l2 = 0.
for mm in range(i, j):
rna[mm].update_energy(list(param))
bp = fil.closing_bp_indices[mm]
l2 += alpha * (actual_bpp[mm] - rna[mm].get_log_bpp(bp[0], bp[1]))**2
prior = guess - param
l2 += w * np.dot(prior, prior)
return l2
0.45
]
Sequence = np.loadtxt("experimental/R101_Sequence.txt", dtype='string')
closing_bp_indices = np.loadtxt("experimental/R101_closing_bp_indices.txt",
dtype='int')
ms2_hairpin_basepairs = np.loadtxt(
"experimental/R101_ms2_hairpin_basepairs.txt", dtype='int')
closingBP_kd = []
fullHP_kd = []
n = len(Sequence)
for mm in range(n):
bp = closing_bp_indices[mm]
closingBP_kd.append(10. / tf.RNA(Sequence[mm], False, energies, True,
False).get_bpp(bp[0], bp[1]))
hp = ms2_hairpin_basepairs[mm]
prob = 1.
for ii in range(7):
prob *= tf.RNA(Sequence[mm], False, energies, True,
False).get_bpp(hp[2 * ii], hp[2 * ii + 1])
fullHP_kd.append(10. / prob)
print mm
#kd.append(np.log(1e-9) - tf.RNA(fil.Sequence[i], False, energies, True, False).get_log_bpp(bp[0],bp[1]))
#log_exp_kd.append(np.log(fil.KDnoligand[i]))
np.savetxt("synthetic/R101_closing_basepair_KD.txt",
closingBP_kd,
delimiter='\t')
np.savetxt("synthetic/R101_full_ms2_KD.txt",
fullHP_kd,
delimiter='\t',
import tinyfold as tf
import scipy.optimize as so
import matplotlib.pyplot as plt
rna = []
energies = [5.69, 6., 4.09, -7.09]
actual_bpp = []
w = 0.01
f = open("sequences_train.txt", 'r')
b = f.readline()
i = 0
data = 0.
while b:
rna.append(tf.RNA(b, False, energies, True, True))
data += len(b)**2
actual_bpp.append(np.array(rna[i].get_bpp_full()))
actual_bpp[i][np.isinf(actual_bpp[i])] = 0.
b = f.readline()
i += 1
f.close()
#guess = np.array([5., 4., 4., -6.])
guess = np.arange(-20, -16)
def cost(param, i, j):
#param = np.array([p, 6., 4.09, -7.09])
l2 = 0.
for mm in range(i, j):
'''R101 DATA'''
energies = np.array([])
#energies = np.array([-0.93, -1.1, -1.33, -2.08, -2.11, -2.24, -2.35, -2.36, -3.26, -3.42, 4.09, 0.45])
'''BPP TRAINING DATA'''
num_training_examples = 100
actual_bpp = [
] # storing log(bpp) for closing stem of hairpin for each input sequence
trouble_sequences = []
sequences = Sequence[:
num_training_examples] #fil.Sequence[:num_training_examples]
for mm in range(num_training_examples):
bp = closing_bp_indices[mm] #fil.closing_bp_indices[i]
rna.append(
tf.RNA(sequences[mm], False, list(energies), True,
False).get_bpp(bp[0], bp[1]))
actual_bpp.append(vienna_closing_bpp[mm])
#rna.append(np.log(10.) - tf.RNA(sequences[mm], False, list(energies), True, False).get_log_bpp(bp[0], bp[1]))
#actual_bpp.append(np.log(vienna_closing_bpp[mm]))
#ms2_prob = 1.
#bp = ms2_hairpin_basepairs[mm]
#for ii in range(7):
# ms2_prob *= tf.RNA(sequences[mm], False, list(energies), True, False).get_bpp(bp[2*ii], bp[2*ii + 1])
#rna.append(ms2_prob)
#if ((ms2_prob < 1e-5) and (vienna_ms2_bpp[mm] > 1e-5)):
# trouble_sequences.append(sequences[mm])
#actual_bpp.append(vienna_ms2_bpp[mm])
#rna.append(np.log(ms2_prob))
#actual_bpp.append(np.log(vienna_ms2_bpp[mm]))
#rna.append(ms2_prob)
#actual_bpp.append(10./KDnoligand[mm])