def test_read_transfac_alphabet_superset(self):
with data_stream("transfac_matrix.txt") as f:
Motif.read_transfac(f, alphabet='TCGA')
# Supplied alphabet can be superset of defacto alphabet.
# Reverts to defacto alphabet
with data_stream("transfac_matrix.txt") as f:
Motif.read_transfac(f, alphabet='TCGAXYZ')
def test_reverse(self):
f = data_stream("transfac_matrix.txt")
m = Motif.read_transfac(f)
f2 = data_stream("transfac_matrix.txt")
m2 = Motif.read_transfac(f2)
m2.reverse()
(K, N) = np.shape(m2)
for k in range(0, K):
for n in range(0, N):
assert (m[k, n] == m2[K - k - 1, n])
f.close()
f2.close()
def test_reverse_complement(self):
f = data_stream("transfac_matrix.txt")
m = Motif.read_transfac(f)
f2 = data_stream("transfac_matrix.txt")
m2 = Motif.read_transfac(f2)
m.complement()
m.reverse()
m2.reverse_complement()
assert (m.array == m2.array).all()
f.close()
f2.close()
def test_complement(self):
f = data_stream("transfac_matrix.txt")
m = Motif.read_transfac(f)
f2 = data_stream("transfac_matrix.txt")
m2 = Motif.read_transfac(f2)
m2.complement()
(K, N) = np.shape(m2)
for k in range(0, K):
assert (m[k, 'A'] == m2[k, 'T'])
assert (m[k, 'G'] == m2[k, 'C'])
assert (m[k, 'C'] == m2[k, 'G'])
assert (m[k, 'T'] == m2[k, 'A'])
f.close()
f2.close()
def test_complement(self):
f = data_stream("transfac_matrix.txt")
m = Motif.read_transfac(f)
f2 = data_stream("transfac_matrix.txt")
m2 = Motif.read_transfac(f2)
m2.complement()
(K, N) = np.shape(m2)
for k in range(0, K):
assert m[k, "A"] == m2[k, "T"]
assert m[k, "G"] == m2[k, "C"]
assert m[k, "C"] == m2[k, "G"]
assert m[k, "T"] == m2[k, "A"]
f.close()
f2.close()
def test_reindex(self):
f = data_stream("transfac_matrix.txt")
m = Motif.read_transfac(f)
f.close()
m2 = m.reindex("TCGA")
assert (str(m2.alphabet) == "TCGA")
for k in range(0, 12):
for i, a in enumerate("AGCT"):
assert m[k, a] == m2[k, a]
def test_read_transfac(self):
f = data_stream("transfac_matrix.txt")
m = Motif.read_transfac(f)
f.close()
assert m[3, 'A'] == 0.0
assert m[0, 'G'] == 2.0
assert np.shape(m.array) == (12, 4)
f.close()
f = data_stream("transfac_matrix2.txt")
m = Motif.read_transfac(f)
f.close()
assert m[3, 'A'] == 3.0
assert m[0, 'G'] == 152.0
assert np.shape(m.array) == (15, 4)
# this one has extra Ps on start of each line
f = data_stream("transfac_matrix3.txt")
m = Motif.read_transfac(f)
f.close()
def get_weblogos(args):
"""Build standard weblogos per convolutional filter."""
# create output directory
if not os.path.exists(args.out_dir):
os.makedirs(args.out_dir)
samples = np.load(args.train_data, mmap_mode='r')
gc_content = np.sum(np.mean(np.mean(samples, axis=1), axis=0)[1:3])
# for each convolutional filter
for file in os.listdir(args.in_dir):
if bool(re.search("_motifs_filter_[0-9]+.*" + args.file_ext, file)) and \
os.stat(args.in_dir + "/" + file).st_size > 0:
c_filter = re.search("filter_[0-9]+", file).group()
filter_index = c_filter.replace("filter_", "")
print("Processing filter: " + filter_index)
fin = open(args.in_dir + "/" + file)
# load motifs from fasta file
if args.file_ext == ".fasta":
seqs = read_seq_data(fin)
prior = parse_prior(str(gc_content), seqs.alphabet)
try:
data = LogoData.from_seqs(seqs, prior)
except ValueError as err:
print(err)
continue
except RuntimeError as err:
print(err)
continue
# load count matrix from transfac file
elif args.file_ext == ".transfac":
motif = Motif.read_transfac(fin)
prior = parse_prior(str(gc_content), motif.alphabet)
try:
data = LogoData.from_counts(motif.alphabet, motif, prior)
except ValueError as err:
print(err)
continue
except RuntimeError as err:
print(err)
continue
# set logo options
options = LogoOptions()
options.logo_title = "filter " + filter_index
options.color_scheme = classic
options.stack_width = std_sizes["large"]
options.resolution = 300
# save filter logo
l_format = LogoFormat(data, options)
png = png_formatter(data, l_format)
with open(
args.out_dir + "/weblogo_" +
file.replace(args.file_ext, ".png"), 'wb') as out_file:
out_file.write(png)
eps = eps_formatter(data, l_format)
with open(
args.out_dir + "/weblogo_" +
file.replace(args.file_ext, ".eps"), 'wb') as out_file:
out_file.write(eps)
def get_weblogos_ext(args):
""" Build extended weblogos per convolutional filter with nucleotide coloring."""
s_max = 1 / args.gain
s_min = -s_max
samples = np.load(args.train_data, mmap_mode='r')
gc_content = np.sum(np.mean(np.mean(samples, axis=1), axis=0)[1:3])
at_content = 1 - gc_content
base_pseudocounts = np.array(
[at_content, gc_content, gc_content, at_content]) / 2.0
# create output directory
if not os.path.exists(args.out_dir):
os.makedirs(args.out_dir)
letter_dict = dict({'A': 0, 'C': 1, 'G': 2, 'T': 3})
# nucleotide color scheme: blue - grey - red
colormap = plt.cm.coolwarm
# for each convolutional filter
for file_fasta in os.listdir(args.fasta_dir):
if bool(re.search("_motifs_filter_[0-9]+.*" + ".fasta", file_fasta)) and \
os.stat(args.fasta_dir + "/" + file_fasta).st_size > 0:
c_filter = re.search("filter_[0-9]+", file_fasta).group()
filter_index = c_filter.replace("filter_", "")
print("Processing filter: " + filter_index)
else:
continue
file_transfac = []
file_scores = [
filename for filename in os.listdir(args.scores_dir) if bool(
re.search(
"rel_filter_" + str(filter_index) +
"_nucleotides\.csv", filename))
]
assert len(
file_scores) < 2, "Multiple score files for filter {}".format(
filter_index)
# load transfac files
if args.logo_dir:
file_transfac = [
filename for filename in os.listdir(args.logo_dir) if bool(
re.search(
"filter_" + str(filter_index) +
"_seq_weighting.transfac", filename))
]
if len(file_transfac) == 0:
continue
assert len(file_transfac
) < 2, "Multiple transfac files for filter {}".format(
filter_index)
# load nucleotide contribution scores
contribution_scores = []
with open(args.scores_dir + "/" + file_scores[0], 'r') as csvfile:
reader = csv.reader(csvfile)
for ind, row in enumerate(reader):
if ind % 2 == 1:
scores = np.array(row, dtype=np.float32)
contribution_scores.append(scores)
# load motifs from fasta file
try:
fin = open(args.fasta_dir + "/" + file_fasta)
seqs = read_seq_data(fin)
except IOError:
print("No data, skipping.")
continue
except ValueError:
print("No data, skipping.")
continue
# load weighted count matrix from transfac file
if args.logo_dir:
fin = open(args.logo_dir + "/" + file_transfac[0])
motif = Motif.read_transfac(fin)
prior = parse_prior(str(gc_content), motif.alphabet)
data = LogoData.from_counts(motif.alphabet, motif, prior)
out_png_name = args.out_dir + "/weblogo_extended_" + file_transfac[
0].replace(".transfac", ".png")
out_eps_name = args.out_dir + "/weblogo_extended_" + file_transfac[
0].replace(".transfac", ".eps")
else:
prior = parse_prior(str(gc_content), seqs.alphabet)
data = LogoData.from_seqs(seqs, prior)
out_png_name = args.out_dir + "/weblogo_extended_" + file_fasta.replace(
".fasta", ".png")
out_eps_name = args.out_dir + "/weblogo_extended_" + file_fasta.replace(
".fasta", ".eps")
seq_names = [seq.name for seq in seqs]
seen = set()
seqs_unique = [
seqs[idx] for idx, seq_name in enumerate(seq_names)
if seq_name not in seen and not seen.add(seq_name)
]
assert len(contribution_scores) == len(
seqs_unique
), "Numbers of contribution scores and sequences differ."
# compute mean contribution score per nucleotide and logo position
mean_scores = np.zeros((len(seqs_unique[0]), len(seqs.alphabet)))
counts = np.zeros_like(data.counts.array)
for r_id, read in enumerate(seqs_unique):
for pos, base in enumerate(read):
base = str(base)
if base in letter_dict.keys():
mean_scores[
pos,
letter_dict[base]] += contribution_scores[r_id][pos]
counts[pos, letter_dict[base]] += 1
# add pseudocount to avoid divion by 0
motif_len = len(seqs_unique[0])
pseudocounts = np.reshape(
np.concatenate([base_pseudocounts] * motif_len, axis=0),
[motif_len, 4])
mean_scores /= (counts + pseudocounts)
# normalize scores to [0, 255] and assign color according the selected color scheme
norm_scores = ((mean_scores - s_min) / (s_max - s_min)) * 255
color_rules = []
for base in letter_dict.keys():
for pos in range(len(seqs[0])):
custom_color = matplotlib.colors.rgb2hex(
colormap(int(norm_scores[pos, letter_dict[base]])))
color_rules.append(SymbolIndexColor(base, [pos], custom_color))
# set logo options
options = LogoOptions()
options.logo_title = "filter " + str(filter_index)
options.color_scheme = ColorScheme(color_rules)
options.stack_width = std_sizes["large"]
options.resolution = 300
# save filter logo
l_format = LogoFormat(data, options)
png = png_formatter(data, l_format)
with open(out_png_name, 'wb') as out_file:
out_file.write(png)
eps = eps_formatter(data, l_format)
with open(out_eps_name, 'wb') as out_file:
out_file.write(eps)