def _plot_motif(self, data, subseqs):
# original structure input was a PWM
if isinstance(subseqs[0], np.ndarray):
rnas, structs = [], []
for pwm in subseqs:
idx = np.argmax(~np.isclose(pwm, 0), axis=1)
rnas.append(''.join(data.alpha_coder.alph0[x] for x in idx //
len(data.alpha_coder.alph1)))
structs.append(
np.zeros((len(rnas[-1]), len(data.alpha_coder.alph1)),
dtype=np.float32))
for i, val in enumerate(idx):
val = val - val % len(data.alpha_coder.alph1)
structs[-1][i] = pwm[i, val:(val +
len(data.alpha_coder.alph1))]
structs = np.sum(structs, 0) / len(structs)
logo_rna = Motif(data.alpha_coder.alph0, sequences=rnas)
logo_struct = Motif(data.alpha_coder.alph1, pwm=structs)
return (logo_rna, logo_struct)
# original structure input was a string
if data.is_rna:
rnas, structs = zip(*(data.alpha_coder.decode(seq)
for seq in subseqs))
logo_rna = Motif(data.alpha_coder.alph0, sequences=rnas)
logo_struct = Motif(data.alpha_coder.alph1, sequences=structs)
return (logo_rna, logo_struct)
# no structure input, just sequence
return Motif(data.one_hot_encoder.alphabet, sequences=subseqs)
def setUp(self):
self.ref_pwm = np.array(
[[0, 0, 0.25, 0], [0.25, 0, 0, 0], [0, 0, 0, 0.25], [
0, 0, 0, 0.25
], [0.25, 0, 0, 0], [0, 0.25, 0, 0], [0.25, 0, 0, 0]],
dtype=np.float32)
self.m = Motif("ACGT", ["GATTACA"])
self.m2 = Motif("ACGT", pwm=self.ref_pwm)
def test_utils_save_as_meme(self):
logos = [Motif('ACGT', ['GATTACA']), Motif('ACGT', ['AAAA'])]
utils.save_as_meme(logos, gettempdir() + "/test.meme")
with open(self.folder + "/data/ref.meme", 'rt') as handle:
ref = handle.read()
with open(gettempdir() + "/test.meme", 'rt') as handle:
comp = handle.read()
self.assertTrue(ref == comp)
remove(gettempdir() + "/test.meme")
def _get_optimized_input(self, model, data, layer_name, node_index, boundary, lr, steps, colors_sequence, colors_structure):
for attempt in range(5):
input_data = np.random.uniform(-boundary, +boundary,
(1, self.params["input_shape"][0], self.params["input_shape"][1]))
input_data, success = self._optimize_input(model, layer_name, node_index, input_data, lr, steps)
if success: break
if not success:
print("Warning: loss did not converge for node {} in layer '{}'".format(node_index, layer_name))
input_data = np.apply_along_axis(utils.softmax, 1, input_data)
if not data.is_rna:
return [Motif(data.one_hot_encoder.alphabet, pwm = input_data).plot(colors_sequence, scale=0.25)]
else:
if data.is_rna_pwm:
annotation_seq = ''.join(x*len(data.alpha_coder.alph1) for x in data.alpha_coder.alph0)
annotation_struct = ''.join(data.alpha_coder.alph1 * len(data.alpha_coder.alph0))
else:
annotation_seq, annotation_struct = data.alpha_coder.decode(data.alpha_coder.alphabet)
pwm_struct = self._extract_pwm(input_data, annotation_struct, data.alpha_coder.alph1)
pwm_seq = self._extract_pwm(input_data, annotation_seq, data.alpha_coder.alph0)
motif_struct = Motif(data.alpha_coder.alph1, pwm = pwm_struct).plot(colors_structure, scale=0.25)
motif_seq = Motif(data.alpha_coder.alph0, pwm = pwm_seq).plot(colors_sequence, scale=0.25)
return [motif_seq, motif_struct]
class Test_Motif(unittest.TestCase):
def setUp(self):
self.ref_pwm = np.array(
[[0, 0, 0.25, 0], [0.25, 0, 0, 0], [0, 0, 0, 0.25], [
0, 0, 0, 0.25
], [0.25, 0, 0, 0], [0, 0.25, 0, 0], [0.25, 0, 0, 0]],
dtype=np.float32)
self.m = Motif("ACGT", ["GATTACA"])
self.m2 = Motif("ACGT", pwm=self.ref_pwm)
def test_motif_init(self):
self.assertTrue(self.m.alphabet == "ACGT")
self.assertTrue(self.m2.alphabet == "ACGT")
def test_motif_valid_pwm(self):
self.assertTrue(self.m.pwm.shape == (7, 4))
self.assertTrue((self.m.pwm >= 0).all() and (self.m.pwm <= 1).all())
self.assertTrue(np.allclose(np.sum(self.m.pwm, axis=1), [1] * 7))
self.assertTrue(self.m2.pwm.shape == (7, 4))
self.assertTrue((self.m2.pwm >= 0).all() and (self.m2.pwm <= 1).all())
self.assertTrue(np.allclose(np.sum(self.m2.pwm, axis=1), [1] * 7))
self.assertTrue(np.allclose(self.m.pwm, self.m2.pwm))
def test_motif_valid_entropies(self):
self.assertTrue(self.m.entropies.shape == (7, ))
self.assertTrue((self.m.entropies >= 0).all()
and (self.m.entropies <= 2).all())
self.assertTrue(self.m2.entropies.shape == (7, ))
self.assertTrue((self.m2.entropies >= 0).all()
and (self.m2.entropies <= 2).all())
self.assertTrue(np.allclose(self.m.entropies, self.m2.entropies))
def test_motif_plot(self):
self.assertTrue(isinstance(self.m.plot(), Image.Image))
self.assertTrue(isinstance(self.m2.plot(), Image.Image))
from random import choice
import gzip
from pysster.Motif import Motif
def rand_dna(length):
return "".join(choice("ACGT") for x in range(length))
num = 5000
with gzip.open("artifical_pos.fasta.gz", "wt") as handle:
seqs = [
rand_dna(20) + "CCCCCCCCCC" + rand_dna(20) + "GGGGGGGGGG" +
rand_dna(80) for x in range(num)
]
for x in range(num):
handle.write(">1\n{}\n".format(seqs[x]))
Motif("ACGT", seqs).plot().save("pos_half1.png")
seqs = [
rand_dna(80) + "AAAAAAAAAA" + rand_dna(20) + "TTTTTTTTTT" +
rand_dna(20) for x in range(num)
]
for x in range(num):
handle.write(">1\n{}\n".format(seqs[x]))
Motif("ACGT", seqs).plot().save("pos_half2.png")
with gzip.open("artifical_neg.fasta.gz", "wt") as handle:
seqs = [rand_dna(140) for x in range(num * 2)]
Motif("ACGT", seqs).plot().save("neg.png")
for x in range(num * 2):
handle.write(">1\n{}\n".format(seqs[x]))
