def get_composition(group, kmer_len, species):
"""
count the composition of each of the segments in a given k-mer and add it to the Counter
"""
chrom, data = group
total_triplets = Counter()
for i, row in data.iterrows():
try:
seq = 0
if kmer_len == 5:
seq = refseq(species, chrom,
int(row['s']) - 2,
int(row['distance']) + 4)
elif kmer_len == 7:
seq = refseq(species, chrom,
int(row['s']) - 3,
int(row['distance']) + 6)
elif kmer_len == 3:
seq = refseq(species, chrom,
int(row['s']) - 1,
int(row['distance']) + 2)
if len(seq) > 0:
total_triplets = total_triplets + Counter(
kmers_generator(seq, kmer_len))
except:
continue
return total_triplets
def obtain_context(df, cont, cent):
try:
seq = refseq('saccer3', df['Chromosome'],
df['Peak genomic coordinate'] - cent, cont)
except:
seq = '-'
return seq
def get_sequence_window(nucid, species, window, kmers=3):
"""Get the sequence for a particular window (including flankig positions)"""
chrom, start, end = nucid.split('_')
# this is already one position before the real start
start_real = int(start)-(window-1)/2
t_win = 0
if kmers == 3:
t_win = window+2
seq = refseq(species, chrom, int(start_real), t_win)
elif kmers == 5:
t_win = window+4
seq = refseq(species, chrom, int(start_real) - 1, t_win)
elif kmers == 7:
t_win = window+6
seq = refseq(species, chrom, int(start_real) - 2, t_win)
# this is to avoid problems if we reach the end of the chromosome, unlikely in hg19 but can happen in yeast
if len(seq) != t_win:
diff = t_win-len(seq)
seq = seq+('N'*diff)
return seq
def get_mutation(row, genome, size):
ref = row['REF']
chr = row['CHR']
pos = int(row['POS'])
try:
kmer = refseq(genome, chr, pos - 2, 5)
except (ValueError, RuntimeError): # out of chrom size, unknown chr
return None
if len(kmer) == 5 and (ref is '-' or kmer[2] == ref) and all(
n in 'ACGT' for n in kmer):
if size == 5:
return kmer
else: # 3-mer by default
return kmer[1:4]
else:
return None
def analyze_experiment(vcf_file):
df = vcf_reader(vcf_file)
df['POS-1'] = df['POS'] - 1
# remove non canonical chromosomes
df['CAN'] = df['CHROM'].apply(lambda x: 'RMV' if 'CACT' in x else 'PASS')
df = df[df['CAN'] == 'PASS']
df['TRIPLET'] = df.apply(
lambda x: refseq('leish', x['CHROM'], x['POS-1'], 3, release=None),
axis=1)
# select whether we have SNVs or others
df['len_alt'] = df['ALT'].str.len()
# number of characters in ref
df['len_ref'] = df['REF'].str.len()
# first classification between SNV and others
df['TYPE'] = df.apply(lambda x: 'SNV'
if (x['len_alt'] == 1) & (x['len_ref'] == 1) and
(x['ALT'] != '-') and (x['REF'] != '-') else 'INDEL',
axis=1)
df = df[df['TYPE'] == 'SNV']
df['VARIANT_CLASS'] = df.apply(create_snv_class, axis=1)
# get whether the mutation has happened in the WT or treated
df['FILTER_WT'] = df['ERR174230.bam'].apply(lambda x: 'RES'
if '0/0' in x else 'WT')
# select only variants in the treated
dic_variants = df[df['FILTER_WT'] ==
'RES']['VARIANT_CLASS'].value_counts().to_dict()
do_plot(dic_variants, 'experiment')
# select only variants in the WT
dic_variants = df[df['FILTER_WT'] ==
'WT']['VARIANT_CLASS'].value_counts().to_dict()
do_plot(dic_variants, 'WT')
def proportion_ww(genome, file):
wanted = 'AT'
count_seq = 0
data = collections.defaultdict(int)
with gzip.open(file, 'rt') as infile:
next(infile)
for line in infile:
line_spl = line.rstrip().split('\t')
try:
seq = refseq(genome, line_spl[0], int(line_spl[2]) - 73, 148)
count_seq += 1
for i in range(len(seq)):
if (seq[i] in wanted) & (seq[i + 1] in wanted):
data[i] += 1
except:
continue
return data, count_seq
def IndelsClassifier(row, genome='hg19'):
"""
This will classify indel and dbs events into the PCAWG categories to feed into the extraction.
We are classifying indels based on the position in the genome we belive they have occured.
a) In the case of deletions:
1. if the first letter in ref is the same as in alt, we conclude that the fragment excised is REF[1:].
2. if the first letter in ref differs to the one in alt, we conclude the entire REF has been excised.
This means that when checking sequences, POS should not be included as in 1) since it is also deleted.
b) In the case of insertions:
1.if the first letter in ref is the same as in alt, we conclude that the fragment inserted is ALT[1:]
2.if the first letter in ref differs to the one in alt, we conclude the insertion is the entire REF.
"""
dipyr = ('C', 'T')
comp = {'A': 'T', 'G': 'C'}
complementary = {'A': 'T', 'T': 'A', 'C': 'G', 'G': 'C'}
dbs_list = {
'AC_CA', 'AC_CG', 'AC_CT', 'AC_GA', 'AC_GG', 'AC_GT', 'AC_TA', 'AC_TG',
'AC_TT', 'AT_CA', 'AT_CC', 'AT_CG', 'AT_GA', 'AT_GC', 'AT_TA', 'CC_AA',
'CC_AG', 'CC_AT', 'CC_GA', 'CC_GG', 'CC_GT', 'CC_TA', 'CC_TG', 'CC_TT',
'CG_AT', 'CG_GC', 'CG_GT', 'CG_TA', 'CG_TC', 'CG_TT', 'CT_AA', 'CT_AC',
'CT_AG', 'CT_GA', 'CT_GC', 'CT_GG', 'CT_TA', 'CT_TC', 'CT_TG', 'GC_AA',
'GC_AG', 'GC_AT', 'GC_CA', 'GC_CG', 'GC_TA', 'TA_AT', 'TA_CG', 'TA_CT',
'TA_GC', 'TA_GG', 'TA_GT', 'TC_AA', 'TC_AG', 'TC_AT', 'TC_CA', 'TC_CG',
'TC_CT', 'TC_GA', 'TC_GG', 'TC_GT', 'TG_AA', 'TG_AC', 'TG_AT', 'TG_CA',
'TG_CC', 'TG_CT', 'TG_GA', 'TG_GC', 'TG_GT', 'TT_AA', 'TT_AC', 'TT_AG',
'TT_CA', 'TT_CC', 'TT_CG', 'TT_GA', 'TT_GC', 'TT_GG'
}
# ===================
# DELETION BIG GROUP
# ===================
# Example --> chr1 99072 CT C
# This means you have removed the T and you keep the C
if row['CLASS'] == 'DEL':
first_nuc_REF = row['REF'][0]
first_nuc_ALT = row['ALT'][0]
# this means it is constant
if first_nuc_REF == first_nuc_ALT:
flag_full = True
endpos = len(row['REF'])
size_del = len(row['REF'][1:])
affected_site = row['REF'][1]
affected_seq = row['REF'][1:]
pos = int(row['POS'])
else:
size_del = len(row['REF'])
affected_site = row['REF'][0]
pos = int(row['POS']) - 1
affected_seq = row['REF']
endpos = len(row['REF']) + 1
# FIRST CLASS : one base DELETION
if size_del == 1:
# if in dipyrimidines group
left_sequence = refseq(genome, row['CHROM'], pos - 4, 5)
right_sequence = refseq(genome, row['CHROM'], pos + 2, 5)
# now we check for the repeats
# we will never have count 0!
count_eq = 1
for i in right_sequence:
if i == affected_site:
count_eq += 1
else:
break
for i in left_sequence[::-1]:
if i == affected_site:
count_eq += 1
else:
break
# if we have 5 repetitions, we have this class
if count_eq > 5:
class_in = 'DEL_{}_1_6+'.format(
comp.get(affected_site, affected_site))
# else, we specifie the dipyr and how many repeats we have
else:
class_in = 'DEL_{}_1_{}'.format(
comp.get(affected_site, affected_site),
count_eq,
)
# SECOND CLASS: more than one base deletion
elif size_del > 1:
# len_ref-1 because the chunck excised also contains the nucleotide before
seq1 = refseq(genome, row['CHROM'], pos + endpos, size_del * 5)
# we want 3' of the reverse
seq2 = refseq(genome, row['CHROM'], pos - size_del * 5,
size_del * 5 + 1)[::-1]
count_eq = 1
# split the sequence into bins of the same size of the deleted region
# Right location
splitted = [
seq1[i:i + size_del] for i in range(0, len(seq1), size_del)
]
for i in splitted:
if i == affected_seq:
count_eq += 1
else:
break
# Left location
splitted = [
seq2[i:i + size_del] for i in range(0, len(seq2), size_del)
]
for i in splitted:
if i == affected_seq[::-1]:
count_eq += 1
else:
break
# if the count is equal or greater than 5, we have this class
if count_eq > 5:
# this yields the DEL_repeats_5+_5+
if size_del >= 5:
class_in = 'DEL_repeats_5+_6+'
else:
class_in = 'DEL_repeats_{}_6+'.format(size_del)
else:
# if we have some repeats found and they are less than 5,
# then they belong to the next class
if count_eq > 1:
if size_del >= 5:
class_in = 'DEL_repeats_5+_{}'.format(count_eq)
else:
class_in = 'DEL_repeats_{}_{}'.format(
size_del, count_eq)
# if no full repeat is found, then give opportunity to microhomology
else:
# get sequence on the right
right_seq = refseq(genome, row['CHROM'], pos + endpos,
size_del)
# get sequence on the left
left_seq = refseq(genome, row['CHROM'], pos - size_del,
size_del + 1)
good = 0
# we go down the size of the indel
for i in np.arange(size_del - 1, 0, -1):
# check right side
tocheck = affected_seq[:i]
tocheck_right = right_seq[:i]
if tocheck == tocheck_right:
good = i
break
# check the left side
tocheck = affected_seq[::-1][:i]
tocheck_left = left_seq[::-1][:i]
if tocheck == tocheck_left:
good = i
break
# if microhomology has been detected
if good > 0:
# this yields the DEL_repeats_5+_5+
if good >= 5:
good = '5+'
if size_del >= 5:
class_in = 'DEL_MH_5+_{}'.format(good)
else:
class_in = 'DEL_MH_{}_{}'.format(size_del, good)
# else this means a deletion with 0 repetitions
else:
if size_del >= 5:
class_in = 'DEL_repeats_5+_1' # we put one according to their info
else:
class_in = 'DEL_repeats_{}_1'.format(size_del)
# ===================
# INSERTIONS BIG GROUP
# ===================
elif row['CLASS'] == 'INS':
first_nuc_REF = row['REF'][0]
first_nuc_ALT = row['ALT'][0]
if first_nuc_REF == first_nuc_ALT:
flag_full = True
endpos = len(row['REF'])
size_del = len(row['ALT'][1:])
affected_site = row['ALT'][1]
affected_seq = row['ALT'][1:]
pos = int(row['POS'])
else:
flag_full = False
size_del = len(row['ALT'])
affected_site = row['ALT'][0]
pos = int(row['POS']) - 1
affected_seq = row['ALT']
endpos = len(row['REF']) + 1
# FIRST CLASS : one base INSERTION
if size_del == 1:
# if in dipyrimidines group
# pos+2 because the deletion is mapped just at the beginning of POS!
right_sequence = refseq(genome, row['CHROM'], pos + endpos, 5)
# else we want whathever is on the 3' (because we would reverse it)
# we want 3' of the reversed. This should include the first nucleotide!
left_sequence = refseq(genome, row['CHROM'], pos - 4, 5)
# we will never have count 0!
count_eq = 0
for i in right_sequence:
if i == affected_site:
count_eq += 1
else:
break
for i in left_sequence[::-1]:
if i == affected_site:
count_eq += 1
else:
break
if count_eq >= 5:
class_in = 'INS_{}_1_5+'.format(
comp.get(affected_site, affected_site))
# else, we specifie the dipyr and how many repeats we have
else:
class_in = 'INS_{}_1_{}'.format(
comp.get(affected_site, affected_site),
count_eq,
)
elif size_del > 1:
# len_ref-1 because the chunck excised also contains the nucleotide before
seq1 = refseq(genome, row['CHROM'], pos + endpos, size_del * 5)
seq2 = refseq(genome, row['CHROM'], pos - size_del * 5,
size_del * 5 + 1)[::-1]
count_eq = 0
# split the sequence into bins of the same size of the deleted region
splitted = [
seq1[i:i + size_del] for i in range(0, len(seq1), size_del)
]
for i in splitted:
if i == affected_seq:
count_eq += 1
else:
break
splitted = [
seq2[i:i + size_del] for i in range(0, len(seq2), size_del)
]
for i in splitted:
if i == affected_seq[::-1]:
count_eq += 1
else:
break
# if the count is equal or greater than 5, we have this class
if count_eq >= 5:
# this yields the INS_repeats_5+_5+
if size_del >= 5:
class_in = 'INS_repeats_5+_5+'
else:
class_in = 'INS_repeats_{}_5+'.format(size_del)
else:
if size_del >= 5:
class_in = 'INS_repeats_5+_{}'.format(count_eq)
else:
class_in = 'INS_repeats_{}_{}'.format(size_del, count_eq)
# ===================
# DBS BIG GROUP
# ===================
elif row['CLASS'] == 'DBS':
# When merging the reverse complementary doublet base substitutions classes into one class, 12 of mutation
# classes have no strandness (e.g. CG > AT), resulting in 78 classes of doublet base substitutions
class_in = '{}_{}'.format(row['REF'], row['ALT'])
if class_in not in dbs_list:
# AC>CA GT>TG
class_in = '{}{}_{}{}'.format(complementary[row['REF'][1]],
complementary[row['REF'][0]],
complementary[row['ALT'][1]],
complementary[row['ALT'][0]])
else:
class_in = 'NOTGOOD'
return class_in
def coronavirus(pos, size=1):
return refseq("sarscov2", "0", pos, size)
def simulate_and_analysis(self, item):
"""
Simulate mutations and analyze simulations
Args:
item (tuple): element of analysis data containing element (str), probs_tree (IntervalTree),
conseq_tree (IntervalTree), n_sim (int)
Returns:
element (str): element of analysis
sim_scores_chunk (list): simulated element's results
sim_cluster_chunk (list): simulated cluster's results
"""
element, probs_tree, n_sim, seed = item
sim_scores_chunk = []
sim_cluster_chunk = []
df_simulated_mutations = []
half_window = (self.simulation_window - 1) // 2
nucleot = {'A', 'C', 'G', 'T'}
np.random.seed(seed)
# Simulate mutations
for mutation in self.mutations_d[element]:
# Get coordinates of randomization window
expected_hotspot_begin = mutation.position - half_window
expected_hotspot_end = mutation.position + half_window
if self.simulation_mode == 'region_restricted':
"""
Region restricted mode samples simulated mutations in a window of length l that fits in the genomic
element.
First, it checks that the genomic region where the mutation is going to be simulated is longer or equal
than l.
If this is true, it calculates the expected start and end positions of the simulation window.
If one of them falls outside the genomic element, the window of length l is displaced to fit in the
genomic element. If both expected start and end positions fall outside the genomic region, the
simulation window is trimmed and simulations are performed inside the genomic region.
If the genomic region is smaller than l, the simulation window becomes the genomic region. This means
that the simulation window is trimmed and simulations are performed between the end and the start of
the genomic region.
"""
if (mutation.region[1] -
mutation.region[0]) >= self.simulation_window:
# Check if hospot outside region
check_5 = expected_hotspot_begin < mutation.region[0]
check_3 = expected_hotspot_end > (mutation.region[1] - 1)
if check_5 and check_3:
hotspot_begin = mutation.region[0]
hotspot_end = mutation.region[
1] - 1 # regions end +1 in tree
elif check_5:
hotspot_begin = mutation.region[0]
hotspot_end = mutation.region[
0] + self.simulation_window - 1 # window //2 per side
elif check_3:
hotspot_end = mutation.region[
1] - 1 # regions end +1 in tree
hotspot_begin = hotspot_end - self.simulation_window + 1 # window //2 per side
else:
hotspot_begin = expected_hotspot_begin
hotspot_end = expected_hotspot_end
else:
hotspot_begin = mutation.region[0]
hotspot_end = mutation.region[
1] - 1 # regions end +1 in tree
else:
"""
Simulations are `mutation centered`, they are centered in the mutated position and can fall outside
the genomic region.
"""
hotspot_begin = expected_hotspot_begin
hotspot_end = expected_hotspot_end
# Map to index
# 3* accounts for alternates in the array of probabilities
# half_window added in probabilities array
start_index = 3 * (hotspot_begin -
(mutation.region[0] - half_window))
end_index = 3 * (hotspot_end - (mutation.region[0] - half_window) +
1) # +1, range and slice
for interval in probs_tree[mutation.group][
mutation.region[0]]: # unique iteration
simulations = np.random.choice(
range(start_index, end_index),
size=n_sim,
p=self.normalize(element,
interval.data[start_index:end_index]))
# Add info per simulated mutation
list_simulations_per_mutation = []
for count, index in enumerate(simulations):
position = mutation.region[0] - half_window + index // 3
ref_nucleotide = bgr.refseq(self.genome,
self.chromosomes_d[element],
position, 1)
# Calculate sorted alternates and obtain simulated alternated from index
if round(index / 3, 1) == (0.7 + index // 3):
alternate_index = 2
else:
alternate_index = 1 if round(
index / 3, 1) == (0.3 + index // 3) else 0
alternate = sorted(
list(nucleot.difference({ref_nucleotide
})))[alternate_index]
# Simulated mutation
list_simulations_per_mutation.append(
Mutation(position, mutation.region, alternate,
mutation.sample, mutation.group))
df_simulated_mutations.append(list_simulations_per_mutation)
# Start analysis
logger.debug('Start analyzing simulations')
for simulated_mutations in zip(*df_simulated_mutations):
cutoff_clusters, element_score = self.analysis(
element, list(simulated_mutations))
sim_scores_chunk.append(element_score)
for interval in cutoff_clusters:
clusters = interval.data.copy()
for cluster, values in clusters.items():
sim_cluster_chunk.append(values['score'])
return element, sim_scores_chunk, sim_cluster_chunk
def mut_probabilities(self, element):
"""
Generate mutational probabilities per position of an element using the sequence context observed mutational
probabilities calculated from the input cohort/s.
Args:
element (str): element to calculate pre-smoothing
Returns:
probs_tree (IntervalTree): IntervalTree of genomic regions. Length == 3*(genomic + simulation window)
skip (bool): if True skip further analysis
"""
nucleot = {'A', 'C', 'G', 'T'}
probs_tree = defaultdict(IntervalTree)
delta = 1 if self.kmer == 3 else 2
half_window = self.simulation_window // 2
skip = False
# Check signatures dictionaries per group
signatures_d = defaultdict()
for group in self.groups_d[element]:
if os.path.isfile(self.path_pickle):
signature = pickle.load(open(self.path_pickle, "rb"))
try:
signatures_d[group] = signature[group]
except KeyError:
raise Exception(
'Signatures for group {} are missing in signatures dictionary\n'
'Please check signatures file {}'.format(
group, self.path_pickle))
else:
skip = True
if not skip:
# Iterate through genomic regions to get their sequences
sequence = ''
for interval in self.regions_d[element]:
probabilities = defaultdict(list)
expected_length = interval[1] - interval[0] + half_window * 2
start = interval[0] - half_window - delta
size = interval[1] - interval[0] + half_window * 2 + delta * 2
try:
sequence = bgr.refseq(self.genome,
self.chromosomes_d[element], start,
size)
except ValueError as e:
logger.error(e, element, start, size, interval[0],
interval[1])
if sequence:
# Search kmer probabilities
for n in range(delta,
len(sequence) - delta): # start to end
ref_kmer = sequence[n - delta:n + delta + 1]
prob = defaultdict(list)
if ref_kmer.count('N') == 0:
# calculate mutational prob to any other kmer
# sort alternates to keep track
for alt in sorted(
list(
nucleot.difference(
{ref_kmer[self.kmer // 2]}))):
for group, signature in signatures_d.items():
prob[group].append(
signature.get(
'{}>{}'.format(ref_kmer, alt), 0))
else:
logger.warning(
'Mutational probabilities for position {0} could not be calculated. '
'Reverting {0}>ALT probabilities to 0'.format(
n))
for group, signature in signatures_d.items():
prob[group].extend([0, 0, 0])
# Extend position info
for group in signatures_d.keys():
probabilities[group].extend(prob[group])
# Check and add
for group in signatures_d.keys():
if sum(probabilities[group]) != 0 and len(
probabilities[group]) == 3 * expected_length:
probs_tree[group].addi(interval[0], interval[1],
probabilities[group])
elif sum(probabilities[group]) == 0:
logger.critical(
'Context based mutational probabilities in {} '
'region {}-{} equal to 0\n'.format(
element, interval[0], interval[1]))
skip = True
break
elif len(probabilities[group]) != 3 * expected_length:
logger.warning(
'{} probabilities list length is different than expected'
.format(element))
skip = True
break
if skip:
break
else:
skip = True
break
if skip:
logger.critical(
'Context based mutational probabilities could not be calculated for {0}\n'
'{0} analysis is skipped'.format(element))
return probs_tree, skip
def get_full_composition(chrom, kmer_len, species):
"""
count the composition of each of the segments and add it to the Counter
"""
seq = refseq(species, chrom, 1, -1)
return Counter(kmers_generator(seq, kmer_len))
