def merge_seqs(seq1, seq2, overlap):
'''Merge seq1 and seq2 assuming there are 'overlap' bases in common.
Aligns the last 'overlap' bases of seq1 with the first 'overlap' bases of
seq2, and if they align well enough, it joins them halfway.
Args:
seq1 (string): first sequence, goes on the left
seq2 (string): second sequence, on the right
overlap (int): usually 1000
Returns:
Joined sequence
'''
# assumes the last "overlap" bases of seq1 and the first "overlap" bases of seq2 overlap
i0 = -overlap
i1 = overlap
if len(seq1) < overlap:
i0 = 0
if len(seq2) < overlap:
i1 = len(seq2)-1
acc,inds = swalign(seq1[i0:],seq2[:i1])
# put some constraints on the accuracy
if acc < 0.70:
raise Exception('Insufficient accuracy for overlap')
# now put the break point halfway between
inds = [x for x in inds if x[0]>0 and x[1]>0]
imid = inds[int(len(inds)/2)]
i0 += imid[0]
i1 = imid[1]
return seq1[:i0] + seq2[i1:]
def Mutate(fastafile, bamfile, fast5dir, region=None, params={}, verbose=0, test=False, reps=4):
"""Run consensus-calling mutations given required info.
This function is the main one called by the setuptools entry points for
consensus calling, that is,
main() -> consensus() -> Mutate(...) -> PSAlign.Mutate(...)
Use this function if you want to basically run poreseq consensus from
within Python.
It takes care of loading events and mutating in different ways, and if the test
flag is specified, it starts with a low-accuracy 2D sequence instead of the ref.
Args:
fastafile (string): reference fasta file (alignment reference)
bamfile (string): BAM-formatted file of alignments to reference
fast5dir (string): folder where fast5 files are contained
region (string): region string, or None for whole reference
params (param dict): parameters to use for loading
verbose (int): 0 for silent, 1 for steps, 2 for full mutations
test (boolean): start with low-accuracy 2D sequence?
reps (int): number of iterations for mutation and refinement
Returns:
tuple(sequence (string), acc (float)): mutated higher-accuracy consensus sequence
along with its accuracy relative to the reference
"""
if 'verbose' not in params:
params['verbose'] = 0
pa = LoadAlignedEvents(fastafile,bamfile,fast5dir,RegionInfo(region),params)
# and the loaded reference sequence
refseq = pa.sequence
# test automatically sets verbose
if test and verbose == 0:
verbose = 1
# we know our algorithm doesn't do great for 1 or 2 events
# in which case we can just shortcut and return the starting seq
if len(pa.events) < 5:
if verbose > 0:
sys.stderr.write("Coverage is 1 or 2, not mutating...\n")
return (refseq, 100)
if verbose > 0:
sys.stderr.write("Mutating {} bases using {} events\n".format(len(refseq),len(pa.events)))
# if test mode, pick a sequence from event sequences
if test:
seq = ""
for ev in pa.events:
pairs = poreseqcpp.swalign(ev.sequence,refseq)[1]
if pairs[-1][1]-pairs[0][1] > len(seq):
seq = ev.sequence[pairs[0][0]:pairs[-1][0]]
pa.sequence = seq
if test:
sys.stderr.write("Starting accuracy: " + str(round(poreseqcpp.swalign(pa.sequence,refseq)[0],1)) + "%\n")
pa.Mutate(reps=reps)
if verbose>0:
acc = poreseqcpp.swalign(pa.sequence,refseq)[0]
sys.stderr.write("Accuracy: " + str(round(acc,1)) + "%\n")
for i in range(reps):
pa.Mutate(seqs='viterbi')
nbases = pa.Refine()
if verbose>0:
acc = poreseqcpp.swalign(pa.sequence,refseq)[0]
sys.stderr.write("Accuracy: " + str(round(acc,1)) + "%\n")
if nbases == 0:
break
# trim ends as requested
if 'end_trim' in params and len(pa.sequence) > 2*params['end_trim']:
pa.sequence = pa.sequence[int(params['end_trim']):-int(params['end_trim'])]
# find the aligned sequence stats
acc,inds = poreseqcpp.swalign(pa.sequence,refseq)
if verbose>0:
errs = np.sum(np.array(inds)==0,0)
sys.stderr.write("Final accuracy: " + str(round(acc,1)) + "%\n")
sys.stderr.write("Insertions: {}, Deletions: {}\n".format(errs[0],errs[1]))
sys.stderr.write("Final coverage: " + str(round(np.mean(pa.Coverage()),1)) + "X\n")
return (pa.sequence, acc)
def Mutate(fastafile,
bamfile,
fast5dir,
region=None,
params={},
verbose=0,
test=False,
reps=4):
"""Run consensus-calling mutations given required info.
This function is the main one called by the setuptools entry points for
consensus calling, that is,
main() -> consensus() -> Mutate(...) -> PSAlign.Mutate(...)
Use this function if you want to basically run poreseq consensus from
within Python.
It takes care of loading events and mutating in different ways, and if the test
flag is specified, it starts with a low-accuracy 2D sequence instead of the ref.
Args:
fastafile (string): reference fasta file (alignment reference)
bamfile (string): BAM-formatted file of alignments to reference
fast5dir (string): folder where fast5 files are contained
region (string): region string, or None for whole reference
params (param dict): parameters to use for loading
verbose (int): 0 for silent, 1 for steps, 2 for full mutations
test (boolean): start with low-accuracy 2D sequence?
reps (int): number of iterations for mutation and refinement
Returns:
tuple(sequence (string), acc (float)): mutated higher-accuracy consensus sequence
along with its accuracy relative to the reference
"""
if 'verbose' not in params:
params['verbose'] = 0
pa = LoadAlignedEvents(fastafile, bamfile, fast5dir, RegionInfo(region),
params)
# and the loaded reference sequence
refseq = pa.sequence
# test automatically sets verbose
if test and verbose == 0:
verbose = 1
# we know our algorithm doesn't do great for 1 or 2 events
# in which case we can just shortcut and return the starting seq
if len(pa.events) < 5:
if verbose > 0:
sys.stderr.write("Coverage is 1 or 2, not mutating...\n")
return (refseq, 100)
if verbose > 0:
sys.stderr.write("Mutating {} bases using {} events\n".format(
len(refseq), len(pa.events)))
# if test mode, pick a sequence from event sequences
if test:
seq = ""
for ev in pa.events:
pairs = poreseqcpp.swalign(ev.sequence, refseq)[1]
if pairs[-1][1] - pairs[0][1] > len(seq):
seq = ev.sequence[pairs[0][0]:pairs[-1][0]]
pa.sequence = seq
if test:
sys.stderr.write(
"Starting accuracy: " +
str(round(poreseqcpp.swalign(pa.sequence, refseq)[0], 1)) + "%\n")
pa.Mutate(reps=reps)
if verbose > 0:
acc = poreseqcpp.swalign(pa.sequence, refseq)[0]
sys.stderr.write("Accuracy: " + str(round(acc, 1)) + "%\n")
for i in range(reps):
pa.Mutate(seqs='viterbi')
nbases = pa.Refine()
if verbose > 0:
acc = poreseqcpp.swalign(pa.sequence, refseq)[0]
sys.stderr.write("Accuracy: " + str(round(acc, 1)) + "%\n")
if nbases == 0:
break
# trim ends as requested
if 'end_trim' in params and len(pa.sequence) > 2 * params['end_trim']:
pa.sequence = pa.sequence[int(params['end_trim']
):-int(params['end_trim'])]
# find the aligned sequence stats
acc, inds = poreseqcpp.swalign(pa.sequence, refseq)
if verbose > 0:
errs = np.sum(np.array(inds) == 0, 0)
sys.stderr.write("Final accuracy: " + str(round(acc, 1)) + "%\n")
sys.stderr.write("Insertions: {}, Deletions: {}\n".format(
errs[0], errs[1]))
sys.stderr.write("Final coverage: " +
str(round(np.mean(pa.Coverage()), 1)) + "X\n")
return (pa.sequence, acc)
