def gene_feature(Y, X, learn_options):
'''
Things like the sequence of the gene, the DNA Tm of the gene, etc.
'''
gene_names = Y['Target gene']
gene_length = np.zeros((gene_names.values.shape[0], 1))
gc_content = np.zeros((gene_names.shape[0], 1))
temperature = np.zeros((gene_names.shape[0], 1))
molecular_weight = np.zeros((gene_names.shape[0], 1))
for gene in gene_names.unique():
seq = util.get_gene_sequence(gene)
gene_length[gene_names.values == gene] = len(seq)
gc_content[gene_names.values == gene] = SeqUtil.GC(seq)
temperature[gene_names.values == gene] = Tm.Tm_staluc(seq, rna=False)
molecular_weight[gene_names.values == gene] = SeqUtil.molecular_weight(
seq, 'DNA')
all = np.concatenate(
(gene_length, gc_content, temperature, molecular_weight), axis=1)
df = pandas.DataFrame(data=all,
index=gene_names.index,
columns=[
'gene length', 'gene GC content',
'gene temperature', 'gene molecular weight'
])
return df
def gene_feature(Y, X, learn_options):
'''
Things like the sequence of the gene, the DNA Tm of the gene, etc.
'''
gene_names = Y['Target gene']
gene_length = np.zeros((gene_names.values.shape[0], 1))
gc_content = np.zeros((gene_names.shape[0], 1))
temperature = np.zeros((gene_names.shape[0], 1))
molecular_weight = np.zeros((gene_names.shape[0], 1))
for gene in gene_names.unique():
seq = util.get_gene_sequence(gene)
gene_length[gene_names.values==gene] = len(seq)
gc_content[gene_names.values==gene] = SeqUtil.GC(seq)
temperature[gene_names.values==gene] = Tm.Tm_staluc(seq, rna=False)
molecular_weight[gene_names.values==gene] = SeqUtil.molecular_weight(seq, 'DNA')
all = np.concatenate((gene_length, gc_content, temperature, molecular_weight), axis=1)
df = pandas.DataFrame(data=all, index=gene_names.index, columns=['gene length',
'gene GC content',
'gene temperature',
'gene molecular weight'])
return df
def gene_feature(Y):
"""
Things like the sequence of the gene, the DNA Tm of the gene, etc.
"""
gene_names = Y["Target gene"]
gene_length = np.zeros((gene_names.values.shape[0], 1))
gc_content = np.zeros((gene_names.shape[0], 1))
temperature = np.zeros((gene_names.shape[0], 1))
molecular_weight = np.zeros((gene_names.shape[0], 1))
for gene in gene_names.unique():
seq = util.get_gene_sequence(gene)
gene_length[gene_names.values == gene] = len(seq)
gc_content[gene_names.values == gene] = SeqUtil.GC(seq)
temperature[gene_names.values == gene] = Tm.Tm_staluc(seq, rna=False)
molecular_weight[gene_names.values == gene] = SeqUtil.molecular_weight(
seq, "DNA")
everything = np.concatenate(
(gene_length, gc_content, temperature, molecular_weight), axis=1)
df = pd.DataFrame(
data=everything,
index=gene_names.index,
columns=[
"gene length",
"gene GC content",
"gene temperature",
"gene molecular weight",
],
)
return df
def local_gene_seq_features(gene_names, learn_options, X):
print(f"building local gene sequence features")
feat = pd.DataFrame(index=X.index)
feat["gene_left_win"] = ""
feat["gene_right_win"] = ""
# number of nulceotides to take to the left and right of the guide
k_mer_length = learn_options["include_gene_guide_feature"]
for gene in gene_names.unique():
gene_seq = Seq.Seq(util.get_gene_sequence(gene)).reverse_complement()
for ps in np.where(gene_names.values == gene)[0]:
guide_seq = Seq.Seq(X["30mer"][ps])
strand = X["Strand"][ps]
if strand == "sense":
guide_seq = guide_seq.reverse_complement()
# figure out the sequence to the left and right of this guide, in the gene
ind = gene_seq.find(guide_seq)
if ind == -1:
if ind == -1:
raise AssertionError("could not find guide in gene")
if gene_seq[ind:(ind + len(guide_seq))] != guide_seq:
raise AssertionError("match not right")
left_win = gene_seq[(ind - k_mer_length):ind]
right_win = gene_seq[(ind + len(guide_seq)):(ind + len(guide_seq) +
k_mer_length)]
if strand == "antisense":
# it's arbitrary which of sense and anti-sense we flip, we just want
# to keep them in the same relative alphabet/direction
left_win = left_win.reverse_complement()
right_win = right_win.reverse_complement()
if left_win.tostring() == "":
raise AssertionError(
f"k_mer_context, {k_mer_length}, is too large")
if len(left_win) != len(right_win):
raise AssertionError(
f"k_mer_context, {k_mer_length}, is too large")
feat.ix[ps, "gene_left_win"] = left_win.tostring()
feat.ix[ps, "gene_right_win"] = right_win.tostring()
print(f"featurizing local context of {gene}")
feature_sets = {}
get_all_order_nuc_features(
feat["gene_left_win"],
feature_sets,
learn_options,
learn_options["order"],
max_index_to_use=maxsize,
prefix="gene_left_win",
)
get_all_order_nuc_features(
feat["gene_right_win"],
feature_sets,
learn_options,
learn_options["order"],
max_index_to_use=maxsize,
prefix="gene_right_win",
)
return feature_sets
def local_gene_seq_features(gene_names, learn_options, X):
print "building local gene sequence features"
feat = pandas.DataFrame(index=X.index)
feat["gene_left_win"] = ""
feat["gene_right_win"] = ""
# number of nulceotides to take to the left and right of the guide
k_mer_length = learn_options['include_gene_guide_feature']
for gene in gene_names.unique():
gene_seq = Seq.Seq(util.get_gene_sequence(gene)).reverse_complement()
for ps in np.where(gene_names.values == gene)[0]:
guide_seq = Seq.Seq(X['30mer'][ps])
strand = X['Strand'][ps]
if strand == 'sense':
guide_seq = guide_seq.reverse_complement()
#gene_seq = gene_seq.reverse_complement()
# figure out the sequence to the left and right of this guide, in the gene
ind = gene_seq.find(guide_seq)
if ind == -1:
#gene_seq = gene_seq.reverse_complement()
#ind = gene_seq.find(guide_seq)
assert ind != -1, "could not find guide in gene"
assert gene_seq[ind:(
ind + len(guide_seq))] == guide_seq, "match not right"
left_win = gene_seq[(ind - k_mer_length):ind]
right_win = gene_seq[(ind + len(guide_seq)):(ind + len(guide_seq) +
k_mer_length)]
if strand == 'antisense':
# it's arbitrary which of sense and anti-sense we flip, we just want
# to keep them in the same relative alphabet/direction
left_win = left_win.reverse_complement()
right_win = right_win.reverse_complement()
assert not left_win.tostring(
) == "", "k_mer_context, %s, is too large" % k_mer_length
assert not left_win.tostring(
) == "", "k_mer_context, %s, is too large" % k_mer_length
assert len(left_win) == len(
right_win), "k_mer_context, %s, is too large" % k_mer_length
feat.ix[ps, "gene_left_win"] = left_win.tostring()
feat.ix[ps, "gene_right_win"] = right_win.tostring()
print "featurizing local context of %s" % (gene)
feature_sets = {}
get_all_order_nuc_features(feat["gene_left_win"],
feature_sets,
learn_options,
learn_options["order"],
max_index_to_use=sys.maxint,
prefix="gene_left_win")
get_all_order_nuc_features(feat["gene_right_win"],
feature_sets,
learn_options,
learn_options["order"],
max_index_to_use=sys.maxint,
prefix="gene_right_win")
return feature_sets
def get_micro_homology_features(gene_names, X):
# originally was flipping the guide itself as necessary, but now flipping the gene instead
print("building microhomology features")
feat = pd.DataFrame(index=X.index)
feat["mh_score"] = ""
feat["oof_score"] = ""
# number of nulceotides to take to the left and right of the guide
K_MER_LENGTH_LEFT = 9
K_MER_LENGTH_RIGHT = 21
for gene in gene_names.unique():
gene_seq = Seq.Seq(util.get_gene_sequence(gene)).reverse_complement()
guide_inds = np.where(gene_names.values == gene)[0]
print(
f"getting microhomology for all {len(guide_inds)} guides in gene {gene}"
)
for ps in guide_inds:
guide_seq = Seq.Seq(X["30mer"][ps])
strand = X["Strand"][ps]
if strand == "sense":
gene_seq = gene_seq.reverse_complement()
# figure out the sequence to the left and right of this guide, in the gene
ind = gene_seq.find(guide_seq)
if ind == -1:
gene_seq = gene_seq.reverse_complement()
ind = gene_seq.find(guide_seq)
else:
pass
if ind == -1:
mh_score = 0
oof_score = 0
else:
if gene_seq[ind:(ind + len(guide_seq))] != guide_seq:
raise AssertionError("match not right")
left_win = gene_seq[(ind - K_MER_LENGTH_LEFT):ind]
right_win = gene_seq[(ind +
len(guide_seq)):(ind + len(guide_seq) +
K_MER_LENGTH_RIGHT)]
if len(left_win.tostring()) != K_MER_LENGTH_LEFT:
raise AssertionError()
if len(right_win.tostring()) != K_MER_LENGTH_RIGHT:
raise AssertionError()
sixtymer = str(left_win) + str(guide_seq) + str(right_win)
if len(sixtymer) != 60:
raise AssertionError("should be of length 60")
mh_score, oof_score = compute_score(sixtymer)
feat.ix[ps, "mh_score"] = mh_score
feat.ix[ps, "oof_score"] = oof_score
print(f"computed microhomology of {str(gene)}")
return pd.DataFrame(feat, dtype="float")
def local_gene_seq_features(gene_names, learn_options, X):
print "building local gene sequence features"
feat = pandas.DataFrame(index=X.index)
feat["gene_left_win"] = ""
feat["gene_right_win"] = ""
# number of nulceotides to take to the left and right of the guide
k_mer_length = learn_options['include_gene_guide_feature']
for gene in gene_names.unique():
gene_seq = Seq.Seq(util.get_gene_sequence(gene)).reverse_complement()
for ps in np.where(gene_names.values==gene)[0]:
guide_seq = Seq.Seq(X['30mer'][ps])
strand = X['Strand'][ps]
if strand=='sense':
guide_seq = guide_seq.reverse_complement()
#gene_seq = gene_seq.reverse_complement()
# figure out the sequence to the left and right of this guide, in the gene
ind = gene_seq.find(guide_seq)
if ind ==-1:
#gene_seq = gene_seq.reverse_complement()
#ind = gene_seq.find(guide_seq)
assert ind != -1, "could not find guide in gene"
assert gene_seq[ind:(ind+len(guide_seq))]==guide_seq, "match not right"
left_win = gene_seq[(ind - k_mer_length):ind]
right_win = gene_seq[(ind + len(guide_seq)):(ind + len(guide_seq) + k_mer_length)]
if strand=='antisense':
# it's arbitrary which of sense and anti-sense we flip, we just want
# to keep them in the same relative alphabet/direction
left_win = left_win.reverse_complement()
right_win = right_win.reverse_complement()
assert not left_win.tostring()=="", "k_mer_context, %s, is too large" % k_mer_length
assert not left_win.tostring()=="", "k_mer_context, %s, is too large" % k_mer_length
assert len(left_win)==len(right_win), "k_mer_context, %s, is too large" % k_mer_length
feat.ix[ps,"gene_left_win"] = left_win.tostring()
feat.ix[ps,"gene_right_win"] = right_win.tostring()
print "featurizing local context of %s" % (gene)
feature_sets = {}
get_all_order_nuc_features(feat["gene_left_win"], feature_sets, learn_options, learn_options["order"], max_index_to_use=sys.maxint, prefix="gene_left_win")
get_all_order_nuc_features(feat["gene_right_win"], feature_sets, learn_options, learn_options["order"], max_index_to_use=sys.maxint, prefix="gene_right_win")
return feature_sets
def get_micro_homology_features(gene_names, learn_options, X):
# originally was flipping the guide itself as necessary, but now flipping the gene instead
print "building microhomology features"
feat = pandas.DataFrame(index=X.index)
feat["mh_score"] = ""
feat["oof_score"] = ""
#with open(r"tmp\V%s_gene_mismatches.csv" % learn_options["V"],'wb') as f:
if True:
# number of nulceotides to take to the left and right of the guide
k_mer_length_left = 9
k_mer_length_right = 21
for gene in gene_names.unique():
gene_seq = Seq.Seq(util.get_gene_sequence(gene)).reverse_complement()
guide_inds = np.where(gene_names.values == gene)[0]
print "getting microhomology for all %d guides in gene %s" % (len(guide_inds), gene)
for j, ps in enumerate(guide_inds):
guide_seq = Seq.Seq(X['30mer'][ps])
strand = X['Strand'][ps]
if strand=='sense':
gene_seq = gene_seq.reverse_complement()
# figure out the sequence to the left and right of this guide, in the gene
ind = gene_seq.find(guide_seq)
if ind==-1:
gene_seq = gene_seq.reverse_complement()
ind = gene_seq.find(guide_seq)
#assert ind != -1, "still didn't work"
#print "shouldn't get here"
else:
#print "all good"
pass
#assert ind != -1, "could not find guide in gene"
if ind==-1:
#print "***could not find guide %s for gene %s" % (str(guide_seq), str(gene))
#if.write(str(gene) + "," + str(guide_seq))
mh_score = 0
oof_score = 0
else:
#print "worked"
assert gene_seq[ind:(ind+len(guide_seq))]==guide_seq, "match not right"
left_win = gene_seq[(ind - k_mer_length_left):ind]
right_win = gene_seq[(ind + len(guide_seq)):(ind + len(guide_seq) + k_mer_length_right)]
#if strand=='antisense':
# # it's arbitrary which of sense and anti-sense we flip, we just want
# # to keep them in the same relative alphabet/direction
# left_win = left_win.reverse_complement()
# right_win = right_win.reverse_complement()
assert len(left_win.tostring())==k_mer_length_left
assert len(right_win.tostring())==k_mer_length_right
sixtymer = str(left_win) + str(guide_seq) + str(right_win)
assert len(sixtymer)==60, "should be of length 60"
mh_score, oof_score = microhomology.compute_score(sixtymer)
feat.ix[ps,"mh_score"] = mh_score
feat.ix[ps,"oof_score"] = oof_score
print "computed microhomology of %s" % (str(gene))
return pandas.DataFrame(feat, dtype='float')
def get_micro_homology_features(gene_names, learn_options, X):
# originally was flipping the guide itself as necessary, but now flipping the gene instead
print "building microhomology features"
feat = pandas.DataFrame(index=X.index)
feat["mh_score"] = ""
feat["oof_score"] = ""
#with open(r"tmp\V%s_gene_mismatches.csv" % learn_options["V"],'wb') as f:
if True:
# number of nulceotides to take to the left and right of the guide
k_mer_length_left = 9
k_mer_length_right = 21
for gene in gene_names.unique():
gene_seq = Seq.Seq(
util.get_gene_sequence(gene)).reverse_complement()
guide_inds = np.where(gene_names.values == gene)[0]
print "getting microhomology for all %d guides in gene %s" % (
len(guide_inds), gene)
for j, ps in enumerate(guide_inds):
guide_seq = Seq.Seq(X['30mer'][ps])
strand = X['Strand'][ps]
if strand == 'sense':
gene_seq = gene_seq.reverse_complement()
# figure out the sequence to the left and right of this guide, in the gene
ind = gene_seq.find(guide_seq)
if ind == -1:
gene_seq = gene_seq.reverse_complement()
ind = gene_seq.find(guide_seq)
#assert ind != -1, "still didn't work"
#print "shouldn't get here"
else:
#print "all good"
pass
#assert ind != -1, "could not find guide in gene"
if ind == -1:
#print "***could not find guide %s for gene %s" % (str(guide_seq), str(gene))
#if.write(str(gene) + "," + str(guide_seq))
mh_score = 0
oof_score = 0
else:
#print "worked"
assert gene_seq[ind:(
ind + len(guide_seq))] == guide_seq, "match not right"
left_win = gene_seq[(ind - k_mer_length_left):ind]
right_win = gene_seq[(ind +
len(guide_seq)):(ind +
len(guide_seq) +
k_mer_length_right)]
#if strand=='antisense':
# # it's arbitrary which of sense and anti-sense we flip, we just want
# # to keep them in the same relative alphabet/direction
# left_win = left_win.reverse_complement()
# right_win = right_win.reverse_complement()
assert len(left_win.tostring()) == k_mer_length_left
assert len(right_win.tostring()) == k_mer_length_right
sixtymer = str(left_win) + str(guide_seq) + str(right_win)
assert len(sixtymer) == 60, "should be of length 60"
mh_score, oof_score = microhomology.compute_score(sixtymer)
feat.ix[ps, "mh_score"] = mh_score
feat.ix[ps, "oof_score"] = oof_score
print "computed microhomology of %s" % (str(gene))
return pandas.DataFrame(feat, dtype='float')
