def test_five_signatures_perturbed(self):
weights = np.zeros((30, ))
weights[3] = .4
weights[7] = .2
weights[13] = .2
weights[17] = .1
weights[23] = .1
tumor_profile, context_counts = generate_tumor_profile(weights)
# Perturb the counts by up to 6% more or less of value
for k, v in context_counts.items():
context_counts[k] = v * (1 + (np.random.rand() * 0.12) - 0.06)
ds = DeconstructSigs(context_counts=context_counts)
# Limit to two deconstructed signatures
reconstructed_weights = ds.which_signatures(5)
self.assertAlmostEqual(reconstructed_weights[3], .4, places=1)
self.assertAlmostEqual(reconstructed_weights[7], .2, places=1)
self.assertAlmostEqual(reconstructed_weights[13], .2, places=1)
self.assertAlmostEqual(reconstructed_weights[17], .1, places=1)
self.assertAlmostEqual(reconstructed_weights[23], .1, places=1)
# Don't impose a signature limit to test that algorithm correctly deduces only five are necessary
reconstructed_weights = ds.which_signatures()
self.assertAlmostEqual(reconstructed_weights[3], .4, places=1)
self.assertAlmostEqual(reconstructed_weights[7], .2, places=1)
self.assertAlmostEqual(reconstructed_weights[13], .2, places=1)
self.assertAlmostEqual(reconstructed_weights[17], .1, places=1)
self.assertAlmostEqual(reconstructed_weights[23], .1, places=1)
def test_five_signatures_discard_insignificant(self):
weights = np.zeros((30, ))
weights[3] = .19
weights[7] = .41
weights[13] = .24
weights[17] = .14
weights[23] = .02
tumor_profile, context_counts = generate_tumor_profile(weights)
ds = DeconstructSigs(context_counts=context_counts)
# Limit to five deconstructed signatures
reconstructed_weights = ds.which_signatures(5)
self.assertAlmostEqual(reconstructed_weights[3], .19, places=2)
self.assertAlmostEqual(reconstructed_weights[7], .41, places=2)
self.assertAlmostEqual(reconstructed_weights[13], .24, places=2)
self.assertAlmostEqual(reconstructed_weights[17], .14, places=2)
self.assertAlmostEqual(reconstructed_weights[23], 0, places=2)
# Don't impose a signature limit to test that algorithm correctly deduces only five necessary and significant
reconstructed_weights = ds.which_signatures()
self.assertAlmostEqual(reconstructed_weights[3], .19, places=2)
self.assertAlmostEqual(reconstructed_weights[7], .41, places=2)
self.assertAlmostEqual(reconstructed_weights[13], .24, places=2)
self.assertAlmostEqual(reconstructed_weights[17], .14, places=2)
self.assertAlmostEqual(reconstructed_weights[23], 0, places=2)
def test_one_signature(self):
weights = np.zeros((30, ))
weights[0] = 1
tumor_profile, context_counts = generate_tumor_profile(weights)
ds = DeconstructSigs(context_counts=context_counts)
# Limit to one deconstructed signatures
reconstructed_weights = ds.which_signatures(1)
self.assertAlmostEqual(reconstructed_weights[0], weights[0], places=2)
# Don't impose a signature limit to test that algorithm correctly deduces only one are necessary
reconstructed_weights = ds.which_signatures()
self.assertAlmostEqual(reconstructed_weights[0], weights[0], places=2)
def test_three_signatures_with_associated(self):
weights = np.zeros((30, ))
weights[3] = .6
weights[11] = .25
weights[27] = .15
tumor_profile, context_counts = generate_tumor_profile(weights)
ds = DeconstructSigs(context_counts=context_counts)
# Limit to two deconstructed signatures
reconstructed_weights = ds.which_signatures(3, associated=[11, 27])
self.assertAlmostEqual(reconstructed_weights[3], 0, places=2)
# Don't impose a signature limit to test that algorithm correctly deduces only three are necessary
reconstructed_weights = ds.which_signatures(associated=[11, 27])
self.assertAlmostEqual(reconstructed_weights[3], 0, places=2)
def test_substitution_standardization(self):
weights = np.zeros((30, ))
weights[0] = 1
tumor_profile, context_counts = generate_tumor_profile(weights)
ds = DeconstructSigs(context_counts=context_counts)
# Standard form means that the ref is a pyrimidine
standardized = ds._DeconstructSigs__standardize_subs('G', 'C')
self.assertEqual(standardized, 'C>G')
standardized = ds._DeconstructSigs__standardize_subs('G', 'A')
self.assertEqual(standardized, 'C>T')
standardized = ds._DeconstructSigs__standardize_subs('G', 'T')
self.assertEqual(standardized, 'C>A')
standardized = ds._DeconstructSigs__standardize_subs('A', 'C')
self.assertEqual(standardized, 'T>G')
standardized = ds._DeconstructSigs__standardize_subs('A', 'G')
self.assertEqual(standardized, 'T>C')
standardized = ds._DeconstructSigs__standardize_subs('A', 'T')
self.assertEqual(standardized, 'T>A')
standardized = ds._DeconstructSigs__standardize_subs('C', 'A')
self.assertEqual(standardized, 'C>A')
standardized = ds._DeconstructSigs__standardize_subs('C', 'G')
self.assertEqual(standardized, 'C>G')
standardized = ds._DeconstructSigs__standardize_subs('C', 'T')
self.assertEqual(standardized, 'C>T')
standardized = ds._DeconstructSigs__standardize_subs('T', 'A')
self.assertEqual(standardized, 'T>A')
standardized = ds._DeconstructSigs__standardize_subs('T', 'C')
self.assertEqual(standardized, 'T>C')
standardized = ds._DeconstructSigs__standardize_subs('T', 'G')
self.assertEqual(standardized, 'T>G')
def test_four_signatures(self):
weights = np.zeros((30, ))
weights[6] = .43
weights[10] = .27
weights[25] = .20
weights[29] = .1
tumor_profile, context_counts = generate_tumor_profile(weights)
ds = DeconstructSigs(context_counts=context_counts)
# Limit to four deconstructed signatures
reconstructed_weights = ds.which_signatures(4)
self.assertAlmostEqual(reconstructed_weights[6], .43, places=2)
self.assertAlmostEqual(reconstructed_weights[10], .27, places=2)
self.assertAlmostEqual(reconstructed_weights[25], .20, places=2)
self.assertAlmostEqual(reconstructed_weights[29], .1, places=2)
# Don't impose a signature limit to test that algorithm correctly deduces only fiour are necessary
reconstructed_weights = ds.which_signatures()
self.assertAlmostEqual(reconstructed_weights[6], .43, places=2)
self.assertAlmostEqual(reconstructed_weights[10], .27, places=2)
self.assertAlmostEqual(reconstructed_weights[25], .20, places=2)
self.assertAlmostEqual(reconstructed_weights[29], .1, places=2)
def test_trinucleotide_standardization(self):
weights = np.zeros((30, ))
weights[0] = 1
tumor_profile, context_counts = generate_tumor_profile(weights)
ds = DeconstructSigs(context_counts=context_counts)
pair = {'A': 'T', 'C': 'G', 'T': 'A', 'G': 'C'}
# Standard form means that a pyrimidine is at the center position of the trinucleotide string
already_standard_form = [
'ACA',
'ACC',
'ACG',
'ACT',
'CCA',
'CCC',
'CCG',
'CCT',
'GCA',
'GCC',
'GCG',
'GCT',
'TCA',
'TCC',
'TCG',
'TCT',
'ATA',
'ATC',
'ATG',
'ATT',
'CTA',
'CTC',
'CTG',
'CTT',
'GTA',
'GTC',
'GTG',
'GTT',
'TTA',
'TTC',
'TTG',
'TTT',
]
for trinuc in already_standard_form:
standardized_trinuc = ds._DeconstructSigs__standardize_trinuc(
trinuc)
self.assertEqual(trinuc, standardized_trinuc)
# Non-standard form means that a purine is a the center position of the trinucleotide string
non_standard_form = [
'{}{}{}'.format(pair[trinuc[2]], pair[trinuc[1]], pair[trinuc[0]])
for trinuc in already_standard_form
]
for i, trinuc in enumerate(non_standard_form):
standardized_trinuc = ds._DeconstructSigs__standardize_trinuc(
trinuc)
self.assertEqual(already_standard_form[i], standardized_trinuc)
# Test that lower case trinucleotide strings get uppercased as well
for trinuc in already_standard_form:
standardized_trinuc = ds._DeconstructSigs__standardize_trinuc(
trinuc.lower())
self.assertEqual(trinuc, standardized_trinuc)
for i, trinuc in enumerate(non_standard_form):
standardized_trinuc = ds._DeconstructSigs__standardize_trinuc(
trinuc.lower())
self.assertEqual(already_standard_form[i], standardized_trinuc)