def simulate_heterodimer_grammar(motif1, motif2, seq_length, min_spacing,
max_spacing, num_pos, num_neg, GC_fraction):
"""
Simulates two classes of sequences with motif1 and motif2:
- Positive class sequences with motif1 and motif2 positioned
min_spacing and max_spacing
- Negative class sequences with independent motif1 and motif2 positioned
anywhere in the sequence, not as a heterodimer grammar
Parameters
----------
seq_length : int, length of sequence
GC_fraction : float, GC fraction in background sequence
num_pos : int, number of positive class sequences
num_neg : int, number of negatice class sequences
motif1 : str, encode motif name
motif2 : str, encode motif name
min_spacing : int, minimum inter motif spacing
max_spacing : int, maximum inter motif spacing
Returns
-------
sequence_arr : 1darray
Array with sequence strings.
y : 1darray
Array with positive/negative class labels.
embedding_arr: list
List of embedding objects.
"""
import simdna
from simdna import synthetic
loaded_motifs = synthetic.LoadedEncodeMotifs(
simdna.ENCODE_MOTIFS_PATH, pseudocountProb=0.001)
motif1_generator = synthetic.ReverseComplementWrapper(
synthetic.PwmSamplerFromLoadedMotifs(loaded_motifs, motif1))
motif2_generator = synthetic.ReverseComplementWrapper(
synthetic.PwmSamplerFromLoadedMotifs(loaded_motifs, motif2))
separation_generator = synthetic.UniformIntegerGenerator(
min_spacing, max_spacing)
embedder = synthetic.EmbeddableEmbedder(
synthetic.PairEmbeddableGenerator(motif1_generator, motif2_generator,
separation_generator))
embed_in_background = synthetic.EmbedInABackground(
synthetic.ZeroOrderBackgroundGenerator(
seq_length, discreteDistribution=get_distribution(GC_fraction)),
[embedder])
generated_sequences = tuple(
synthetic.GenerateSequenceNTimes(embed_in_background,
num_pos).generateSequences())
grammar_sequence_arr = np.array(
[generated_seq.seq for generated_seq in generated_sequences])
positive_embedding_arr = [
generated_seq.embeddings for generated_seq in generated_sequences
]
nongrammar_sequence_arr, _, negative_embedding_arr = simulate_multi_motif_embedding(
[motif1, motif2], seq_length, 2, 2, num_neg, GC_fraction)
sequence_arr = np.concatenate((grammar_sequence_arr, nongrammar_sequence_arr))
y = np.array([[True]] * num_pos + [[False]] * num_neg)
embedding_arr = positive_embedding_arr + negative_embedding_arr
return sequence_arr, y, embedding_arr
def test_simple_motif_grammar(self):
seq_len = 100
min_sep = 2
max_sep = 6
random.seed(1234)
np.random.seed(1234)
num_sequences = 4000
loaded_motifs = sn.LoadedEncodeMotifs(
simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=0.001)
motif1_generator = sn.PwmSamplerFromLoadedMotifs(
loaded_motifs, "SIX5_known5")
motif2_generator = sn.PwmSamplerFromLoadedMotifs(
loaded_motifs, "ZNF143_known2")
separation_generator = sn.UniformIntegerGenerator(min_sep,max_sep)
embedder = sn.EmbeddableEmbedder(
sn.PairEmbeddableGenerator(
motif1_generator, motif2_generator, separation_generator))
embed_in_background = sn.EmbedInABackground(
sn.ZeroOrderBackgroundGenerator(seq_len),
[embedder])
generated_sequences = sn.GenerateSequenceNTimes(
embed_in_background, num_sequences).generateSequences()
generated_seqs = [seq for seq in generated_sequences]
separations = defaultdict(lambda: 0)
for seq in generated_seqs:
assert len(seq.seq) == seq_len
embedding1 = seq.embeddings[0]
embedding2 = seq.embeddings[1]
embedding3 = seq.embeddings[2]
assert len(embedding1.what) == len(embedding1.what.string)
assert len(embedding2.what) == len(embedding2.what.string)
assert len(embedding3.what) == (len(embedding1.what)+
len(embedding2.what)+
embedding3.what.separation)
#testing that the string of the first motif is placed correctly
assert (seq.seq[
embedding1.startPos:embedding1.startPos+len(embedding1.what)]
== embedding1.what.string)
#testing that the string of the second motif is placed correctly
assert (seq.seq[
embedding2.startPos:embedding2.startPos+len(embedding2.what)]
== embedding2.what.string)
#testing that the motifs are placed correctly
assert ((embedding2.startPos - (embedding1.startPos
+ len(embedding1.what.string)))
== embedding3.what.separation)
#test separation is within the right limits
assert embedding3.what.separation >= min_sep
assert embedding3.what.separation <= max_sep
#log the separation; will later test distribution
separations[embedding3.what.separation] += 1
for possible_sep in range(min_sep, max_sep+1):
np.testing.assert_almost_equal(
separations[possible_sep]/float(num_sequences),
1.0/(max_sep-min_sep+1),2)
def test_multi_motif_embedding(self):
motif_names = ["CTCF_known1", "IRF_known1",
"SPI1_known1", "CTCF_known2", "CTCF_disc1"]
loaded_motifs = sn.LoadedEncodeMotifs(simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=0.001)
position_generator = sn.UniformPositionGenerator()
embedders = [sn.SubstringEmbedder(sn.PwmSamplerFromLoadedMotifs(
loaded_motifs, motif_name),
position_generator, name=motif_name)
for motif_name in motif_names]
min_selected_motifs = 1
max_selected_motifs = 4
quantity_generator = sn.UniformIntegerGenerator(min_selected_motifs,
max_selected_motifs)
combined_embedder = [sn.RandomSubsetOfEmbedders(
quantity_generator, embedders)]
embed_in_background = sn.EmbedInABackground(
sn.ZeroOrderBackgroundGenerator(
300, discreteDistribution={'A':0.3, 'C':0.2, 'G':0.2, 'T':0.3}),
combined_embedder)
generated_sequences = tuple(sn.GenerateSequenceNTimes(
embed_in_background, 8000).generateSequences())
sequence_arr = np.array([generated_seq.seq for
generated_seq in generated_sequences])
label_generator = sn.IsInTraceLabelGenerator(np.array(motif_names))
y = np.array([label_generator.generateLabels(generated_seq)
for generated_seq in generated_sequences]).astype(bool)
embedding_arr = [generated_seq.embeddings for generated_seq in generated_sequences]
num_embeddings_count = defaultdict(lambda: 0)
for seq, labels, embeddings, generated_seq in zip(sequence_arr, y, embedding_arr, generated_sequences):
motifs_embedded = set()
num_embeddings_count[len(embeddings)] += 1
for embedding in embeddings:
#assert that the string selected is correct
assert embedding.what.string ==\
seq[embedding.startPos:
(embedding.startPos+len(embedding.what.string))]
motifs_embedded.add(embedding.what.getDescription())
assert len(motifs_embedded) == len(embeddings) #non-redundant
for (motif_idx, motif_name) in enumerate(motif_names):
if motif_name in motifs_embedded:
assert labels[motif_idx]==True
else:
assert labels[motif_idx]==False
#assert that the num selected is drawn correctly from a uniform dist
for num_selected_motifs in range(min_selected_motifs,
max_selected_motifs+1):
np.testing.assert_almost_equal(
num_embeddings_count[num_selected_motifs]/float(len(sequence_arr)),
1.0/(max_selected_motifs-min_selected_motifs+1),2)
#there also shouldn't be a preference for any one motif over others
np.testing.assert_almost_equal(
np.sum(y,axis=0)/float(np.sum(y)), 1.0/len(motif_names), 2)
def simple_motif_embedding(motif_name, seq_length, num_seqs, GC_fraction):
"""
Simulates sequences with a motif embedded anywhere in the sequence.
Parameters
----------
motif_name : str
encode motif name
seq_length : int
length of sequence
num_seqs: int
number of sequences
GC_fraction : float
GC basepair fraction in background sequence
Returns
-------
sequence_arr : 1darray
Array with sequence strings.
embedding_arr: 1darray
Array of embedding objects.
"""
import simdna
from simdna import synthetic
if motif_name is None:
embedders = []
else:
loaded_motifs = synthetic.LoadedEncodeMotifs(simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=0.001)
substring_generator = synthetic.PwmSamplerFromLoadedMotifs(
loaded_motifs, motif_name)
embedders = [
synthetic.SubstringEmbedder(
synthetic.ReverseComplementWrapper(substring_generator))
]
embed_in_background = synthetic.EmbedInABackground(
synthetic.ZeroOrderBackgroundGenerator(
seq_length, discreteDistribution=get_distribution(GC_fraction)),
embedders)
generated_sequences = tuple(
synthetic.GenerateSequenceNTimes(embed_in_background,
num_seqs).generateSequences())
sequence_arr = np.array(
[generated_seq.seq for generated_seq in generated_sequences])
embedding_arr = [
generated_seq.embeddings for generated_seq in generated_sequences
]
return sequence_arr, embedding_arr
def test_central_positions(self):
pseudocount_prob = 0.001
pwm_name = "CTCF_known1"
num_sequences = 10000
sequence_length = 50
loaded_motifs = sn.LoadedEncodeMotifs(simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=pseudocount_prob)
substring_generator = sn.PwmSamplerFromLoadedMotifs(
loaded_motifs, pwm_name)
position_generator = sn.InsideCentralBp(30)
embedders = [
sn.SubstringEmbedder(substring_generator, position_generator)
]
embed_in_background = sn.EmbedInABackground(
sn.ZeroOrderBackgroundGenerator(sequence_length,
discreteDistribution={
'A': 0.3,
'C': 0.2,
'G': 0.2,
'T': 0.3
}), embedders)
generated_sequences = list(
sn.GenerateSequenceNTimes(embed_in_background,
num_sequences).generateSequences())
motif_length = len(loaded_motifs.getPwm(pwm_name).getRows())
start_pos_count = np.zeros(sequence_length - motif_length + 1)
for seq in generated_sequences:
assert len(seq.seq) == sequence_length
embeddings = seq.embeddings
for embedding in embeddings:
assert (embedding.what.string ==
seq.seq[embedding.startPos:embedding.startPos +
len(embedding.what.string)])
start_pos_count[embedding.startPos] += 1
start_pos_count = start_pos_count / float(len(generated_sequences))
#the *1.0 is for conversion to float
expected_start_pos_count = np.zeros_like(start_pos_count).astype(
"float32")
#expect motif to be embedded only in the central 40bp
expected_start_pos_count[10:(40 - motif_length +
1)] = 1.0 / (30.0 - motif_length + 1)
np.testing.assert_almost_equal(start_pos_count,
expected_start_pos_count, 2)
def do(options):
if (options.seed is not None):
import numpy as np
np.random.seed(options.seed)
from simdna import random
random.seed(options.seed)
outputFileName_core = util.addArguments("DensityEmbedding",
[util.ArgumentToAdd(options.prefix, "prefix"),
util.ArrArgument(options.motifNames, "motifs"),
util.ArgumentToAdd(options.min_motifs, "min"),
util.ArgumentToAdd(options.max_motifs, "max"),
util.ArgumentToAdd(options.mean_motifs, "mean"),
util.FloatArgument(options.zero_prob, "zeroProb"),
util.ArgumentToAdd(options.seqLength, "seqLength"),
util.ArgumentToAdd(options.posSdevInBp, "posSdevInBp"),
util.ArgumentToAdd(options.numSeqs, "numSeqs")])
loadedMotifs = synthetic.LoadedEncodeMotifs(options.pathToMotifs, pseudocountProb=0.001)
embedInBackground = synthetic.EmbedInABackground(
backgroundGenerator=synthetic.ZeroOrderBackgroundGenerator(seqLength=options.seqLength)
, embedders=[
synthetic.RepeatedEmbedder(
synthetic.SubstringEmbedder(
synthetic.ReverseComplementWrapper(
substringGenerator=synthetic.PwmSamplerFromLoadedMotifs(
loadedMotifs=loadedMotifs,motifName=motifName),
reverseComplementProb=0.5
),
positionGenerator=
synthetic.positiongen.NormalDistributionPositionGenerator(
stdInBp=options.posSdevInBp)),
quantityGenerator=synthetic.ZeroInflater(synthetic.MinMaxWrapper(
synthetic.PoissonQuantityGenerator(options.mean_motifs),
theMax=options.max_motifs, theMin=options.min_motifs), zeroProb=options.zero_prob)
)
for motifName in options.motifNames
]
)
sequenceSet = synthetic.GenerateSequenceNTimes(embedInBackground, options.numSeqs)
synthetic.printSequences(outputFileName_core+".simdata", sequenceSet,
includeFasta=True, includeEmbeddings=True,
prefix=options.prefix)
def motif_density(motif_name,
seq_length,
num_seqs,
min_counts,
max_counts,
GC_fraction,
central_bp=None):
"""
Returns sequences with motif density, along with embeddings array.
"""
import simdna
from simdna import synthetic
loaded_motifs = synthetic.LoadedEncodeMotifs(simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=0.001)
substring_generator = synthetic.PwmSamplerFromLoadedMotifs(
loaded_motifs, motif_name)
if central_bp is not None:
position_generator = synthetic.InsideCentralBp(central_bp)
else:
position_generator = synthetic.UniformPositionGenerator()
quantity_generator = synthetic.UniformIntegerGenerator(
min_counts, max_counts)
embedders = [
synthetic.RepeatedEmbedder(
synthetic.SubstringEmbedder(
synthetic.ReverseComplementWrapper(substring_generator),
position_generator), quantity_generator)
]
embed_in_background = synthetic.EmbedInABackground(
synthetic.ZeroOrderBackgroundGenerator(
seq_length, discreteDistribution=get_distribution(GC_fraction)),
embedders)
generated_sequences = tuple(
synthetic.GenerateSequenceNTimes(embed_in_background,
num_seqs).generateSequences())
sequence_arr = np.array(
[generated_seq.seq for generated_seq in generated_sequences])
embedding_arr = [
generated_seq.embeddings for generated_seq in generated_sequences
]
return sequence_arr, embedding_arr
def test_uniform_positions(self):
pseudocount_prob = 0.001
pwm_name = "CTCF_known1"
num_sequences = 10000
sequence_length = 50
loaded_motifs = sn.LoadedEncodeMotifs(simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=pseudocount_prob)
substring_generator = sn.PwmSamplerFromLoadedMotifs(
loaded_motifs, pwm_name)
position_generator = sn.UniformPositionGenerator()
embedders = [
sn.SubstringEmbedder(substring_generator, position_generator)
]
embed_in_background = sn.EmbedInABackground(
sn.ZeroOrderBackgroundGenerator(sequence_length,
discreteDistribution={
'A': 0.3,
'C': 0.2,
'G': 0.2,
'T': 0.3
}), embedders)
generated_sequences = list(
sn.GenerateSequenceNTimes(embed_in_background,
num_sequences).generateSequences())
motif_length = len(loaded_motifs.getPwm(pwm_name).getRows())
start_pos_count = np.zeros(sequence_length - motif_length + 1)
for seq in generated_sequences:
assert len(seq.seq) == sequence_length
embeddings = seq.embeddings
for embedding in embeddings:
assert (embedding.what.string ==
seq.seq[embedding.startPos:embedding.startPos +
len(embedding.what.string)])
start_pos_count[embedding.startPos] += 1
start_pos_count = start_pos_count / float(len(generated_sequences))
np.testing.assert_almost_equal(start_pos_count,
1.0 / len(start_pos_count), 2)
def test_density_motif_embedding(self):
random.seed(1234)
np.random.seed(1234)
min_counts = 2
max_counts = 5
pseudocount_prob = 0.001
pwm_name = "CTCF_known1"
num_sequences = 5000
loaded_motifs = sn.LoadedEncodeMotifs(simdna.ENCODE_MOTIFS_PATH,
pseudocountProb=pseudocount_prob)
substring_generator = sn.PwmSamplerFromLoadedMotifs(
loaded_motifs, pwm_name)
position_generator = sn.UniformPositionGenerator()
quantity_generator = sn.UniformIntegerGenerator(min_counts, max_counts)
embedders = [
sn.RepeatedEmbedder(
sn.SubstringEmbedder(
sn.ReverseComplementWrapper(
substring_generator), position_generator),
quantity_generator)]
embed_in_background = sn.EmbedInABackground(
sn.ZeroOrderBackgroundGenerator(
500, discreteDistribution={'A':0.3,'C':0.2,
'G':0.2,'T':0.3}),
embedders)
generated_sequences = list(sn.GenerateSequenceNTimes(
embed_in_background, num_sequences).generateSequences())
assert len(generated_sequences) == num_sequences
actual_pwm = np.array([[0.095290, 0.318729, 0.083242, 0.502738],
[0.182913, 0.158817, 0.453450, 0.204819],
[0.307777, 0.053669, 0.491785, 0.146769],
[0.061336, 0.876232, 0.023001, 0.039430],
[0.008762, 0.989047, 0.000000, 0.002191],
[0.814896, 0.014239, 0.071194, 0.099671],
[0.043812, 0.578313, 0.365827, 0.012048],
[0.117325, 0.474781, 0.052632, 0.355263],
[0.933114, 0.012061, 0.035088, 0.019737],
[0.005488, 0.000000, 0.991218, 0.003293],
[0.365532, 0.003293, 0.621295, 0.009879],
[0.059276, 0.013172, 0.553238, 0.374314],
[0.013187, 0.000000, 0.978022, 0.008791],
[0.061538, 0.008791, 0.851648, 0.078022],
[0.114411, 0.806381, 0.005501, 0.073707],
[0.409241, 0.014301, 0.557756, 0.018702],
[0.090308, 0.530837, 0.338106, 0.040749],
[0.128855, 0.354626, 0.080396, 0.436123],
[0.442731, 0.199339, 0.292952, 0.064978]])
actual_pwm = actual_pwm*(1-pseudocount_prob) + pseudocount_prob/4
np.testing.assert_almost_equal(np.sum(actual_pwm,axis=-1),1.0,6)
np.testing.assert_almost_equal(
actual_pwm,
np.array(loaded_motifs.getPwm(pwm_name).getRows()))
letter_to_index = {'A':0, 'C':1, 'G':2, 'T':3}
reconstructed_pwm_fwd = np.zeros_like(actual_pwm)
reconstructed_pwm_rev = np.zeros_like(actual_pwm)
quantity_distribution = defaultdict(lambda: 0)
total_fwd_embeddings = 0.0
total_rev_embeddings = 0.0
for seq in generated_sequences:
embeddings = seq.embeddings
quantity_distribution[len(embeddings)] += 1
for embedding in embeddings:
assert (embedding.what.string
==seq.seq[embedding.startPos:
embedding.startPos+len(embedding.what.string)])
if ('revComp' in embedding.what.getDescription()):
total_rev_embeddings += 1
else:
total_fwd_embeddings += 1
for char_idx, char in enumerate(embedding.what.string):
if ('revComp' in embedding.what.getDescription()):
arr = reconstructed_pwm_rev
else:
arr = reconstructed_pwm_fwd
arr[char_idx][letter_to_index[char]] += 1
total_embeddings = total_fwd_embeddings + total_rev_embeddings
np.testing.assert_almost_equal(
total_fwd_embeddings/total_embeddings, 0.5, 2)
#normalize each column of reconstructed_pwm
reconstructed_pwm_fwd = reconstructed_pwm_fwd/total_fwd_embeddings
reconstructed_pwm_rev = reconstructed_pwm_rev/total_rev_embeddings
np.testing.assert_almost_equal(actual_pwm, reconstructed_pwm_fwd, 2)
np.testing.assert_almost_equal(actual_pwm,
reconstructed_pwm_rev[::-1,::-1], 2)
#test the quantities of motifs were sampled uniformly
for quantity in range(min_counts, max_counts+1):
np.testing.assert_almost_equal(
quantity_distribution[quantity]/float(num_sequences),
1.0/(max_counts-min_counts+1),2)
def get_embedder(motif_name):
substring_generator = synthetic.PwmSamplerFromLoadedMotifs(
loaded_motifs, motif_name)
return synthetic.SubstringEmbedder(
synthetic.ReverseComplementWrapper(substring_generator),
name=motif_name)