def __init__(self, chromosome_name, bam_file_path, draft_file_path, truth_bam, hp_tag, train_mode):
"""
Initialize a manager object
:param chromosome_name: Name of the chromosome
:param bam_file_path: Path to the BAM file
:param draft_file_path: Path to the reference FASTA file
:param truth_bam: Path to the truth sequence to reference mapping file
"""
# --- initialize handlers ---
# create objects to handle different files and query
self.bam_path = bam_file_path
self.fasta_path = draft_file_path
self.bam_handler = PEPPER_HP.BAM_handler(bam_file_path)
self.fasta_handler = PEPPER_HP.FASTA_handler(draft_file_path)
self.train_mode = train_mode
self.hp_tag = hp_tag
self.downsample_rate = 1.0
self.truth_bam = None
if self.train_mode:
self.truth_bam = truth_bam
# --- initialize names ---
# name of the chromosome
self.chromosome_name = chromosome_name
def __init__(self, reference_file_path, contigs, sample_name, output_dir,
filename):
self.fasta_handler = PEPPER_HP.FASTA_handler(reference_file_path)
self.contigs = contigs
vcf_header = self.get_vcf_header(sample_name, contigs)
self.vcf_file = VariantFile(output_dir + filename + '.vcf',
'w',
header=vcf_header)
def reads_to_reference_realignment(self, region_start, region_end, reads):
# PERFORMS LOCAL REALIGNMENT OF READS TO THE REFERENCE
if not reads:
return []
ref_start = region_start
ref_end = region_end + AlingerOptions.ALIGNMENT_SAFE_BASES
ref_sequence = self.fasta_handler.get_reference_sequence(
self.chromosome_name, ref_start, ref_end)
aligner = PEPPER_HP.ReadAligner(ref_start, ref_end, ref_sequence)
realigned_reads = aligner.align_reads_to_reference(reads)
# generate_pileup_from_reads.pileup_from_reads(ref_sequence, ref_start, ref_end, realigned_reads)
return realigned_reads
def __init__(self, vcf1, vcf2, ref_fasta, only_overlapping=True, discard_phase=False, detailed_info=False):
"""Initialize variant merging.
Merge variants from two haploid VCFs into a diploid vcf. Variants in
one file which overlap with variants in the other will have their alts
padded.
.. warning::
Variants in a single vcf file should not overlap with each other.
:param vcf1, vcf2: paths to haploid vcf files.
:param ref_fasta: path to reference.fasta file.
:param only_overlapping: bool, merge only overlapping variants (not
adjacent ones).
:param discard_phase: bool, if False, preserve phase, else output
unphased variants.
"""
self.only_overlapping = only_overlapping
self.discard_phase = discard_phase
self.detailed_info = detailed_info
self.vcfs = [VCFReader(vcf) for vcf in (vcf1, vcf2)]
for vcf in self.vcfs:
vcf.index() # create tree
self.fasta = pysam.FastaFile(ref_fasta)
all_contigs = list(set(itertools.chain(*[v.chroms for v in self.vcfs])))
all_contigs = sorted(all_contigs, key=natural_key)
fasta_handler = PEPPER_HP.FASTA_handler(ref_fasta)
sqs = fasta_handler.get_chromosome_names()
self.chroms = []
for sq in sqs:
if sq not in all_contigs:
continue
sq_id = sq
ln = fasta_handler.get_chromosome_sequence_length(sq)
self.chroms.append((sq_id, ln))
def create_summary(self, truth_bam, hp_tag, train_mode, realignment_flag):
log_prefix = "[" + self.chromosome_name + ":" + str(self.region_start_position) + "-" \
+ str(self.region_end_position) + "]"
all_images = []
all_labels = []
all_positions = []
all_image_chunk_ids = []
all_ref_seq = []
if train_mode:
# get the reads from the bam file
truth_bam_handler = PEPPER_HP.BAM_handler(truth_bam)
# truth reads
include_truth_supp = True
truth_read_mq = 60
truth_read_baseq = 0
# get the reads from the bam file
truth_reads = truth_bam_handler.get_reads(
self.chromosome_name, self.region_start_position,
self.region_end_position, include_truth_supp, truth_read_mq,
truth_read_baseq)
# do a local realignment of truth reads to reference
if realignment_flag:
truth_reads = self.reads_to_reference_realignment(
self.region_start_position, self.region_end_position,
truth_reads)
truth_regions = []
for read in truth_reads:
# start, end, read, is_kept, is_h1
truth_regions.append([read.pos, read.pos_end - 1, read, True])
# these are all the regions we will use to generate summaries from.
# It's important to notice that we need to realign the reads to the reference before we do that.
truth_regions = self.remove_conflicting_regions(truth_regions)
if not truth_regions:
# sys.stderr.write(TextColor.GREEN + "INFO: " + log_prefix + " NO TRAINING REGION FOUND.\n"
# + TextColor.END)
return [], [], [], [], []
for region in truth_regions:
region_start, region_end, truth_read, is_kept = tuple(region)
if not is_kept:
continue
ref_start = region_start
ref_end = region_end + 1
# ref_seq should contain region_end_position base
ref_seq = self.fasta_handler.get_reference_sequence(
self.chromosome_name, ref_start, ref_end)
read_start = max(0, region_start)
read_end = region_end
all_reads = self.bam_handler.get_reads(
self.chromosome_name, read_start, read_end,
ReadFilterOptions.INCLUDE_SUPPLEMENTARY,
ReadFilterOptions.MIN_MAPQ, ReadFilterOptions.MIN_BASEQ)
total_reads = len(all_reads)
if total_reads == 0:
continue
if total_reads > AlingerOptions.MAX_READS_IN_REGION:
# https://github.com/google/nucleus/blob/master/nucleus/util/utils.py
# reservoir_sample method utilized here
random = np.random.RandomState(AlingerOptions.RANDOM_SEED)
sample = []
for i, read in enumerate(all_reads):
if len(sample) < AlingerOptions.MAX_READS_IN_REGION:
sample.append(read)
else:
j = random.randint(0, i + 1)
if j < AlingerOptions.MAX_READS_IN_REGION:
sample[j] = read
all_reads = sample
# sys.stderr.write(TextColor.GREEN + "INFO: " + log_prefix + " TOTAL " + str(total_reads)
# + " READS FOUND.\n" + TextColor.END)
start_time = time.time()
if realignment_flag:
all_reads = self.reads_to_reference_realignment(
read_start, read_end, all_reads)
# sys.stderr.write(TextColor.GREEN + "INFO: " + log_prefix + " REALIGNMENT OF TOTAL "
# + str(total_reads) + " READS TOOK: " + str(round(time.time()-start_time, 5))
# + " secs\n" + TextColor.END)
summary_generator = PEPPER_HP.SummaryGenerator(
ref_seq, self.chromosome_name, ref_start, ref_end)
summary_generator.generate_train_summary(
all_reads, region_start, region_end, truth_read, hp_tag)
image_summary = summary_generator.chunk_image_train(
ImageSizeOptions.SEQ_LENGTH, ImageSizeOptions.SEQ_OVERLAP,
ImageSizeOptions.IMAGE_HEIGHT)
all_images.extend(image_summary.images)
all_labels.extend(image_summary.labels)
all_positions.extend(image_summary.positions)
all_image_chunk_ids.extend(image_summary.chunk_ids)
all_ref_seq.extend(image_summary.refs)
else:
# HERE REALIGN THE READS TO THE REFERENCE THEN GENERATE THE SUMMARY TO GET A POLISHED HAPLOTYPE
read_start = max(0, self.region_start_position)
read_end = self.region_end_position
all_reads = self.bam_handler.get_reads(
self.chromosome_name, read_start, read_end,
ReadFilterOptions.INCLUDE_SUPPLEMENTARY,
ReadFilterOptions.MIN_MAPQ, ReadFilterOptions.MIN_BASEQ)
total_reads = len(all_reads)
if total_reads == 0:
return [], [], [], [], []
if total_reads > AlingerOptions.MAX_READS_IN_REGION:
# https://github.com/google/nucleus/blob/master/nucleus/util/utils.py
# reservoir_sample method utilized here
random = np.random.RandomState(AlingerOptions.RANDOM_SEED)
sample = []
for i, read in enumerate(all_reads):
if len(sample) < AlingerOptions.MAX_READS_IN_REGION:
sample.append(read)
else:
j = random.randint(0, i + 1)
if j < AlingerOptions.MAX_READS_IN_REGION:
sample[j] = read
all_reads = sample
# sys.stderr.write(TextColor.PURPLE + "INFO: " + log_prefix + " TOTAL " + str(total_reads) + " READS FOUND\n"
# + TextColor.END)
if realignment_flag:
start_time = time.time()
all_reads = self.reads_to_reference_realignment(
self.region_start_position, self.region_end_position,
all_reads)
# sys.stderr.write(TextColor.GREEN + "INFO: " + log_prefix + " REALIGNMENT OF TOTAL " + str(total_reads)
# + " READS TOOK: " + str(round(time.time()-start_time, 5)) + " secs\n" + TextColor.END)
# ref_seq should contain region_end_position base
ref_seq = self.fasta_handler.get_reference_sequence(
self.chromosome_name, self.region_start_position,
self.region_end_position + 1)
summary_generator = PEPPER_HP.SummaryGenerator(
ref_seq, self.chromosome_name, self.region_start_position,
self.region_end_position)
summary_generator.generate_summary(all_reads,
self.region_start_position,
self.region_end_position,
hp_tag)
image_summary = summary_generator.chunk_image(
ImageSizeOptions.SEQ_LENGTH, ImageSizeOptions.SEQ_OVERLAP,
ImageSizeOptions.IMAGE_HEIGHT)
all_images.extend(image_summary.images)
all_labels.extend(image_summary.labels)
all_positions.extend(image_summary.positions)
all_image_chunk_ids.extend(image_summary.chunk_ids)
all_ref_seq.extend(image_summary.refs)
assert (len(all_images) == len(all_labels) == len(all_image_chunk_ids)
== len(all_ref_seq))
return all_images, all_labels, all_positions, all_image_chunk_ids, all_ref_seq
def call_variant(bam_filepath, fasta_filepath, output_dir, threads, region,
model_path, batch_size, gpu_mode, callers_per_gpu, device_ids,
num_workers, sample_name):
"""
Run all the sub-modules to polish an input assembly.
"""
start_time = time.time()
# check the bam file
if not os.path.isfile(bam_filepath) or not PEPPER_HP.BAM_handler(
bam_filepath):
sys.stderr.write("ERROR: CAN NOT LOCATE BAM FILE.\n")
exit(1)
# check the fasta file
if not os.path.isfile(fasta_filepath):
sys.stderr.write("ERROR: CAN NOT LOCATE FASTA FILE.\n")
exit(1)
# check the model file
if not os.path.isfile(model_path):
sys.stderr.write("ERROR: CAN NOT LOCATE MODEL FILE.\n")
exit(1)
# check number of threads
if threads <= 0:
sys.stderr.write("ERROR: THREAD NEEDS TO BE >=0.\n")
exit(1)
# check batch_size
if batch_size <= 0:
sys.stderr.write("ERROR: batch_size NEEDS TO BE >0.\n")
exit(1)
# check num_workers
if num_workers < 0:
sys.stderr.write("ERROR: num_workers NEEDS TO BE >=0.\n")
exit(1)
# check if gpu inference can be done
if gpu_mode:
if not torch.cuda.is_available():
sys.stderr.write("ERROR: TORCH IS NOT BUILT WITH CUDA.\n")
sys.stderr.write(
"SEE TORCH CAPABILITY:\n$ python3\n"
">>> import torch \n"
">>> torch.cuda.is_available()\n If true then cuda is avilable"
)
exit(1)
# check if all devices are available
if device_ids is not None:
device_ids = [int(i) for i in device_ids.split(',')]
for device_id in device_ids:
major_capable, minor_capable = torch.cuda.get_device_capability(
device=device_id)
if major_capable < 0:
sys.stderr.write("ERROR: GPU DEVICE: " + str(device_id) +
" IS NOT CUDA CAPABLE.\n")
sys.stderr.write(
"Try running: $ python3\n"
">>> import torch \n"
">>> torch.cuda.get_device_capability(device=" +
str(device_id) + ")\n")
exit(1)
else:
sys.stderr.write("INFO: CAPABILITY OF GPU#" + str(device_id) +
":\t" + str(major_capable) + "-" +
str(minor_capable) + "\n")
timestr = time.strftime("%m%d%Y_%H%M%S")
output_dir = UserInterfaceSupport.handle_output_directory(output_dir)
image_output_directory_hp1 = output_dir + "images_" + str(
timestr) + "/hp1_images/"
image_output_directory_hp2 = output_dir + "images_" + str(
timestr) + "/hp2_images/"
prediction_output_directory_hp1 = output_dir + "predictions_" + str(
timestr) + "/hp1/"
prediction_output_directory_hp2 = output_dir + "predictions_" + str(
timestr) + "/hp2/"
candidate_output_directory_hp1 = output_dir + "candidate_variants_" + str(
timestr) + "/hp1/"
candidate_output_directory_hp2 = output_dir + "candidate_variants_" + str(
timestr) + "/hp2/"
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: RUN-ID: " + str(timestr) + "\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: IMAGE OUTPUT: " +
str(image_output_directory_hp1) + "\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 1.1: GENERATING IMAGES FOR HAPLOTYPE 1\n")
make_images(bam_filepath, fasta_filepath, region,
image_output_directory_hp1, 1, threads)
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: RUN-ID: " + str(timestr) + "\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: IMAGE OUTPUT: " +
str(image_output_directory_hp2) + "\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 1.2: GENERATING IMAGES FOR BOTH HAPLOTYPE 2\n")
make_images(bam_filepath, fasta_filepath, region,
image_output_directory_hp2, 2, threads)
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 2.1: RUNNING INFERENCE ON HAPLOTYPE 1\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: OUTPUT: " +
str(prediction_output_directory_hp1) + "\n")
run_inference(image_output_directory_hp1, model_path, batch_size,
num_workers, prediction_output_directory_hp1, device_ids,
callers_per_gpu, gpu_mode, threads)
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 2.2: RUNNING INFERENCE ON HAPLOTYPE 2\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: OUTPUT: " +
str(prediction_output_directory_hp1) + "\n")
run_inference(image_output_directory_hp2, model_path, batch_size,
num_workers, prediction_output_directory_hp2, device_ids,
callers_per_gpu, gpu_mode, threads)
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 3.1: CALLING VARIANTS ON HAPLOTYPE 1\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: OUTPUT: " + str(candidate_output_directory_hp1) +
"\n")
process_candidates(prediction_output_directory_hp1, fasta_filepath,
sample_name, candidate_output_directory_hp1, threads)
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 3.2: CALLING VARIANTS ON HAPLOTYPE 2\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: OUTPUT: " + str(candidate_output_directory_hp2) +
"\n")
process_candidates(prediction_output_directory_hp2, fasta_filepath,
sample_name, candidate_output_directory_hp2, threads)
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] STEP 4: MERGING VARIANTS.\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) +
"] INFO: OUTPUT: " + str(output_dir) + "\n")
haploid2diploid(
candidate_output_directory_hp1 + 'candidates_as_variants.vcf',
candidate_output_directory_hp2 + 'candidates_as_variants.vcf',
fasta_filepath, output_dir)
end_time = time.time()
mins = int((end_time - start_time) / 60)
secs = int((end_time - start_time)) % 60
sys.stderr.write("[" + datetime.now().strftime('%m-%d-%Y %H:%M:%S') +
"] TOTAL ELAPSED TIME FOR VARIANT CALLING: " + str(mins) +
" Min " + str(secs) + " Sec\n")
def get_chromosome_list(chromosome_names, ref_file, bam_file, region_bed):
"""
PARSES THROUGH THE CHROMOSOME PARAMETER TO FIND OUT WHICH REGIONS TO PROCESS
:param chromosome_names: NAME OF CHROMOSOME
:param ref_file: PATH TO THE REFERENCE FILE
:param bam_file: PATH TO BAM FILE
:return: LIST OF CHROMOSOME IN REGION SPECIFIC FORMAT
"""
if not chromosome_names and not region_bed:
fasta_handler = PEPPER_HP.FASTA_handler(ref_file)
bam_handler = PEPPER_HP.BAM_handler(bam_file)
bam_contigs = bam_handler.get_chromosome_sequence_names()
fasta_contigs = fasta_handler.get_chromosome_names()
common_contigs = list(set(fasta_contigs) & set(bam_contigs))
common_contigs = list(set(common_contigs) - set(EXCLUDED_HUMAN_CONTIGS))
if len(common_contigs) == 0:
sys.stderr.write("[" + datetime.now().strftime('%m-%d-%Y %H:%M:%S') + "] "
+ "ERROR: NO COMMON CONTIGS FOUND BETWEEN THE BAM FILE AND THE FASTA FILE.")
sys.stderr.flush()
exit(1)
common_contigs = sorted(common_contigs, key=UserInterfaceSupport.natural_key)
sys.stderr.write("[" + datetime.now().strftime('%m-%d-%Y %H:%M:%S') + "] INFO: COMMON CONTIGS FOUND: " + str(common_contigs) + "\n")
sys.stderr.flush()
chromosome_name_list = []
for contig_name in common_contigs:
chromosome_name_list.append((contig_name, None))
return chromosome_name_list
if region_bed:
chromosome_name_list = []
with open(region_bed) as fp:
line = fp.readline()
cnt = 1
while line:
line_to_list = line.rstrip().split('\t')
chr_name, start_pos, end_pos = line_to_list[0], int(line_to_list[1]), int(line_to_list[2])
region = sorted([start_pos, end_pos])
chromosome_name_list.append((chr_name, region))
line = fp.readline()
cnt += 1
return chromosome_name_list
split_names = chromosome_names.strip().split(',')
split_names = [name.strip() for name in split_names]
chromosome_name_list = []
for name in split_names:
# split on region
region = None
if ':' in name:
name_region = name.strip().split(':')
if len(name_region) != 2:
sys.stderr.write("ERROR: --region INVALID value.\n")
exit(0)
name, region = tuple(name_region)
region = region.strip().split('-')
region = [int(pos) for pos in region]
if len(region) != 2 or not region[0] <= region[1]:
sys.stderr.write("ERROR: --region INVALID value.\n")
exit(0)
range_split = name.split('-')
if len(range_split) > 1:
chr_prefix = ''
for p in name:
if p.isdigit():
break
else:
chr_prefix = chr_prefix + p
int_ranges = []
for item in range_split:
s = ''.join(i for i in item if i.isdigit())
int_ranges.append(int(s))
int_ranges = sorted(int_ranges)
for chr_seq in range(int_ranges[0], int_ranges[-1] + 1):
chromosome_name_list.append((chr_prefix + str(chr_seq), region))
else:
chromosome_name_list.append((name, region))
return chromosome_name_list
def chromosome_level_parallelization(chr_list,
bam_file,
draft_file,
truth_bam,
hp_tag,
output_path,
total_threads,
train_mode,
realignment_flag):
if train_mode:
max_size = 10000
else:
max_size = 10000
start_time = time.time()
fasta_handler = PEPPER_HP.FASTA_handler(draft_file)
all_intervals = []
# first calculate all the intervals that we need to process
for chr_name, region in chr_list:
# contig update message
if not region:
interval_start, interval_end = (0, fasta_handler.get_chromosome_sequence_length(chr_name) - 1)
else:
interval_start, interval_end = tuple(region)
interval_start = max(0, interval_start)
interval_end = min(interval_end, fasta_handler.get_chromosome_sequence_length(chr_name) - 1)
# this is the interval size each of the process is going to get which is 10^6
# I will split this into 10^4 size inside the worker process
for pos in range(interval_start, interval_end, max_size):
pos_start = max(interval_start, pos - ImageSizeOptions.MIN_IMAGE_OVERLAP)
pos_end = min(interval_end, pos + max_size + ImageSizeOptions.MIN_IMAGE_OVERLAP)
all_intervals.append((chr_name, pos_start, pos_end))
# all intervals calculated now
# contig update message
sys.stderr.write("[" + datetime.now().strftime('%m-%d-%Y %H:%M:%S') + "] "
+ "INFO: TOTAL CONTIGS: " + str(len(chr_list))
+ " TOTAL INTERVALS: " + str(len(all_intervals)) + "\n")
sys.stderr.flush()
args = (output_path, bam_file, draft_file, truth_bam, hp_tag, train_mode, realignment_flag)
with concurrent.futures.ProcessPoolExecutor(max_workers=total_threads) as executor:
futures = [executor.submit(UserInterfaceSupport.image_generator, args, all_intervals, total_threads,
thread_id)
for thread_id in range(0, total_threads)]
for fut in concurrent.futures.as_completed(futures):
if fut.exception() is None:
# get the results
thread_id = fut.result()
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) + "] INFO: THREAD "
+ str(thread_id) + " FINISHED SUCCESSFULLY.\n")
else:
sys.stderr.write("ERROR: " + str(fut.exception()) + "\n")
fut._result = None # python issue 27144
end_time = time.time()
mins = int((end_time - start_time) / 60)
secs = int((end_time - start_time)) % 60
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) + "] INFO: FINISHED IMAGE GENERATION\n")
sys.stderr.write("[" + str(datetime.now().strftime('%m-%d-%Y %H:%M:%S')) + "] INFO: ELAPSED TIME: " + str(mins) + " Min " + str(secs) + " Sec\n")