def CheckAlignments(peptide_sequences, query_token, other_tokens):
"""check wether query aligns to all others.
"""
if param_loglevel >= 3:
print "# checking query %s and sbjcts %s" % (query_token,
str(other_tokens))
sys.stdout.flush()
if query_token not in peptide_sequences:
return True
result = alignlib_lite.makeAlignmentVector()
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -1.0)
row_seq = alignlib_lite.makeSequence(peptide_sequences[query_token])
for x in other_tokens:
if x not in peptide_sequences:
continue
col_seq = alignlib_lite.makeSequence(peptide_sequences[x])
alignator.align(result, row_seq, col_seq)
if param_loglevel >= 5:
print "# %s - %s = %f" % (query_token, x, result.getScore())
if result.getScore() > param_min_alignment_score:
return True
return False
def CheckAlignments(peptide_sequences, query_token, other_tokens):
"""check wether query aligns to all others.
"""
if param_loglevel >= 3:
print "# checking query %s and sbjcts %s" % (query_token, str(other_tokens))
sys.stdout.flush()
if query_token not in peptide_sequences:
return True
result = alignlib_lite.makeAlignmentVector()
alignator = alignlib_lite.makeAlignatorDPFull(alignlib_lite.ALIGNMENT_LOCAL,
-10.0, -1.0)
row_seq = alignlib_lite.makeSequence(peptide_sequences[query_token])
for x in other_tokens:
if x not in peptide_sequences:
continue
col_seq = alignlib_lite.makeSequence(peptide_sequences[x])
alignator.align(result, row_seq, col_seq)
if param_loglevel >= 5:
print "# %s - %s = %f" % (query_token, x, result.getScore())
if result.getScore() > param_min_alignment_score:
return True
return False
def FilterConflicts(old_predictions, new_predictions, removed_predictions,
min_overlap, peptide_sequences):
"""remove conflicts.
Remove overlapping entries between different queries.
Only remove those sequences, which are alignable.
If they are alignable, take the sequence with the highest score and highest coverage.
(Take both, if score and coverage are not correlated.)
"""
##################################################################################################
## sort predictions by genomic region
if isinstance(old_predictions, PredictionFile.PredictionFile):
old_predictions.sort(('mSbjctToken', 'mSbjctStrand',
'mSbjctGenomeFrom', 'mSbjctGenomeTo'))
else:
old_predictions.sort(lambda x, y: cmp(
(x.mSbjctToken, x.mSbjctStrand, x.mSbjctGenomeFrom, x.
mSbjctGenomeTo), (y.mSbjctToken, y.mSbjctStrand, y.
mSbjctGenomeFrom, y.mSbjctGenomeTo)))
##################################################################################################
## filter predictions and resolve conflicts based on genomic overlap
## deleted segments are put in a temporary storage space.
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, param_gop, param_gep)
result = alignlib_lite.makeAlignmentVector()
alignments = {}
noverlaps = 0
nredundants = 0
nnew = 0
last_prediction = None
for this_prediction in old_predictions:
try:
this_query_peptide, this_query_status, this_query_gene, this_query_transcript = \
re.split("\s+", this_prediction.mQueryToken)
except ValueError:
this_query_gene = None
if not last_prediction:
last_prediction = this_prediction
last_query_gene = this_query_gene
continue
overlap = min(last_prediction.mSbjctGenomeTo, this_prediction.mSbjctGenomeTo) - \
max(last_prediction.mSbjctGenomeFrom, this_prediction.mSbjctGenomeFrom)
union = max(last_prediction.mSbjctGenomeTo, this_prediction.mSbjctGenomeTo) - \
min(last_prediction.mSbjctGenomeFrom, this_prediction.mSbjctGenomeFrom)
# resolve overlap between different genes
if overlap > 0 and \
(last_query_gene != this_query_gene or last_query_gene == None):
noverlaps += 1
relative_overlap = 100 * overlap / union
# Start conflict resolution, if overlap is above threshold.
# Keep higher scoring segment.
#
# Check if queries are homologous.
if relative_overlap >= param_max_percent_overlap:
if peptide_sequences:
if last_prediction.mQueryToken < this_prediction.mQueryToken:
key = "%s-%s" % (last_prediction.mQueryToken,
this_prediction.mQueryToken)
else:
key = "%s-%s" % (this_prediction.mQueryToken,
last_prediction.mQueryToken)
if not alignments.has_key(key):
result.clear()
alignator.align(
result,
alignlib_lite.makeSequence(peptide_sequences[
this_prediction.mQueryToken]),
alignlib_lite.makeSequence(peptide_sequences[
last_prediction.mQueryToken]))
alignments[key] = result.getScore()
if result.getScore() >= param_min_score_overlap:
nredundants += 1
if alignments[key] >= param_min_score_overlap:
is_overlap = 1
else:
is_overlap = 0
else:
is_overlap = 1
else:
is_overlap = 0
else:
is_overlap = 0
if is_overlap:
# take best prediction. If difference is very small, set
# difference to 0 (difference does not matter). In this case,
# the first prediction is taken.
d1 = last_prediction.mQueryCoverage - this_prediction.mQueryCoverage
if float(abs(d1)) / float(last_prediction.mQueryCoverage
) < param_conflicts_min_difference:
d1 = 0
d2 = last_prediction.score - this_prediction.score
if float(abs(d2)) / float(
this_prediction.score) < param_conflicts_min_difference:
d2 = 0
if d1 >= 0 and d2 >= 0:
if param_loglevel >= 2:
print "# CONFLICT: kept %i(%s-%i), overlap=%i(%5.2f), removed: %s" % (
last_prediction.mPredictionId,
last_prediction.mQueryToken,
last_prediction.mSbjctGenomeFrom, overlap,
relative_overlap, str(this_prediction))
if param_benchmarks:
if CheckBenchmark(this_prediction, last_prediction):
print "# BENCHMARK KEPT with overlap=%i(%5.2f): %s" % (
overlap, relative_overlap, str(last_prediction))
removed_predictions.append(this_prediction)
continue
elif d1 <= 0 and d2 <= 0:
if param_loglevel >= 2:
print "# CONFLICT: kept %i(%s-%i), overlap=%i(%5.2f), removed: %s" % (
this_prediction.mPredictionId,
this_prediction.mQueryToken,
this_prediction.mSbjctGenomeFrom, overlap,
relative_overlap, str(last_prediction))
if param_benchmarks:
if CheckBenchmark(last_prediction, this_prediction):
print "# BENCHMARK KEPT with overlap=%i(%5.2f): %s" % (
overlap, relative_overlap, str(this_prediction))
removed_predictions.append(last_prediction)
last_prediction = this_prediction
last_query_gene = this_query_gene
continue
else:
if param_loglevel >= 2:
print "# CONFLICT: non-correlated score/coverage. Keeping both %i(%s-%i) (%5.2f/%i/%i) and %i(%s-%i) (%5.2f/%i/%i)" % \
(this_prediction.mPredictionId,
this_prediction.mQueryToken, this_prediction.mSbjctGenomeFrom,
this_prediction.score, this_prediction.mQueryCoverage,
this_prediction.mPercentIdentity,
last_prediction.mPredictionId,
last_prediction.mQueryToken, last_prediction.mSbjctGenomeFrom,
last_prediction.score, last_prediction.mQueryCoverage,
last_prediction.mPercentIdentity)
new_predictions.append(last_prediction)
nnew += 1
last_query_gene = this_query_gene
last_prediction = this_prediction
new_predictions.append(last_prediction)
nnew += 1
if param_loglevel >= 1:
print "# calculated %i alignments for %i potential conflicts (%i above threshold)" % \
(len(alignments), noverlaps, nredundants)
return nnew
def ProcessRegion(predictions,
region_id,
region,
peptide_sequences=None,
filter_queries={}):
"""process a set of matches to a region.
resolve region according to homology.
"""
if options.loglevel >= 3:
options.stdlog.write(
"###################################################################\n"
)
options.stdlog.write("# resolving %i predictions in region %s\n" %
(len(predictions), str(region)))
sys.stdout.flush()
predictions.sort(lambda x, y: cmp(x.score, y.score))
predictions.reverse()
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, options.gop, options.gep)
result = alignlib_lite.makeAlignmentVector()
cluster = []
map_sequence2cluster = range(0, len(predictions))
edges = []
noutput, nskipped = 0, 0
if peptide_sequences:
for x in range(len(predictions)):
if options.loglevel >= 5:
options.stdlog.write(
"# filtering from %i with prediction %i: %s\n" %
(x, predictions[x].mPredictionId,
predictions[x].mQueryToken))
sys.stdout.flush()
if map_sequence2cluster[x] != x:
continue
region_id += 1
edges = []
if predictions[x].mQueryToken not in filter_queries:
edges.append(predictions[x])
else:
nskipped += 1
for y in range(x + 1, len(predictions)):
if map_sequence2cluster[y] != y:
continue
if predictions[x].mQueryToken < predictions[y].mQueryToken:
key = "%s-%s" % (predictions[x].mQueryToken,
predictions[y].mQueryToken)
else:
key = "%s-%s" % (predictions[y].mQueryToken,
predictions[x].mQueryToken)
# check if predictions are overlapping on the genomic sequence
if min(predictions[x].mSbjctGenomeTo, predictions[y].mSbjctGenomeTo) - \
max(predictions[x].mSbjctGenomeFrom, predictions[y].mSbjctGenomeFrom) < 0:
if options.loglevel >= 4:
options.stdlog.write(
"# alignment of predictions %i and %i: no overlap on genomic sequence, thus skipped\n"
% (predictions[x].mPredictionId,
predictions[y].mPredictionId))
sys.stdout.flush()
continue
if not global_alignments.has_key(key):
seq1 = peptide_sequences[predictions[x].mQueryToken]
seq2 = peptide_sequences[predictions[y].mQueryToken]
result.clear()
s1 = alignlib_lite.makeSequence(seq1)
s2 = alignlib_lite.makeSequence(seq2)
alignator.align(result, s1, s2)
c1 = 100 * \
(result.getRowTo() - result.getRowFrom()) / len(seq1)
c2 = 100 * \
(result.getColTo() - result.getColFrom()) / len(seq2)
min_cov = min(c1, c2)
max_cov = max(c1, c2)
identity = alignlib_lite.calculatePercentIdentity(
result, s1, s2) * 100
# check if predictions overlap and they are homologous
if result.getScore() >= options.overlap_min_score and \
max_cov >= options.overlap_max_coverage and \
min_cov >= options.overlap_min_coverage and \
identity >= options.overlap_min_identity:
global_alignments[key] = True
else:
global_alignments[key] = False
if options.loglevel >= 4:
options.stdlog.write(
"# alignment=%s score=%i pid=%5.2f c1=%i c2=%i min_cov=%i max_cov=%i homolog=%s\n"
% (key, result.getScore(), identity, c1, c2,
min_cov, max_cov, global_alignments[key]))
sys.stdout.flush()
if global_alignments[key]:
map_sequence2cluster[y] = x
if predictions[y].mQueryToken not in filter_queries:
edges.append(predictions[y])
else:
nskipped += 1
noutput += PrintEdges(region_id, region, edges)
return region_id, noutput, nskipped
def EliminateRedundantEntries(
rep, data, eliminated_predictions, options, peptides, extended_peptides, filter_quality=None, this_quality=None
):
"""eliminate redundant entries in a set."""
eliminated = []
rep_id = rep.transcript_id
rep_coverage, rep_pid = rep.mQueryCoverage, rep.mPid
alignator = alignlib_lite.makeAlignatorDPFull(alignlib_lite.ALIGNMENT_LOCAL, options.gop, options.gep)
result = alignlib_lite.makeAlignmentVector()
rep_seq = peptides[rep_id]
rep_extended_seq = extended_peptides[rep_id]
for entry in data:
mem_id, mem_coverage, mem_pid, mem_quality = (
entry.transcript_id,
entry.mQueryCoverage,
entry.mPid,
entry.mQuality,
)
mem_seq = peptides[mem_id]
mem_extended_seq = extended_peptides[mem_id]
if options.loglevel >= 4:
options.stdlog.write("# processing: id=%s class=%s\n" % (mem_id, mem_quality))
if mem_id in eliminated_predictions:
continue
if mem_extended_seq == rep_extended_seq:
eliminated_predictions[mem_id] = rep_id
eliminated.append((mem_id, "i"))
elif mem_extended_seq in rep_extended_seq:
eliminated_predictions[mem_id] = rep_id
eliminated.append((mem_id, "p"))
else:
if mem_quality != this_quality or mem_quality in options.quality_exclude_same:
seq1 = alignlib_lite.makeSequence(str(rep_seq))
seq2 = alignlib_lite.makeSequence(str(mem_seq))
alignator.align(result, seq1, seq2)
if options.loglevel >= 5:
options.stdlog.write("# ali\n%s\n" % alignlib_lite.AlignmentFormatExplicit(result, seq1, seq2))
pidentity = 100 * alignlib_lite.calculatePercentIdentity(result, seq1, seq2)
num_gaps = result.getNumGaps()
if options.loglevel >= 4:
options.stdlog.write(
"# processing: id=%s class=%s pid=%5.2f rep_cov=%i mem_cov=%i\n"
% (mem_id, mem_quality, pidentity, rep_coverage, mem_coverage)
)
if pidentity >= options.min_identity:
keep = False
if rep_coverage < mem_coverage - options.safety_coverage or rep_pid < mem_pid - options.safety_pide:
keep = True
reason = "covpid"
elif num_gaps >= options.max_gaps and mem_coverage > rep_coverage - options.safety_coverage:
keep = True
reason = "gaps"
elif (
mem_coverage >= rep_coverage - options.safety_coverage
and 100 * (result.getColTo() - result.getColFrom()) / len(mem_seq) < options.max_member_coverage
):
keep = True
reason = "memcov"
if keep:
options.stdlog.write(
"# WARNING: not removing possibly good prediction: %s: rep = %s, mem = %s, rep_cov=%i, rep_pid=%i, mem_cov=%i, mem_pid=%i\n"
% (reason, rep_id, mem_id, rep_coverage, rep_pid, mem_coverage, mem_pid)
)
else:
eliminated_predictions[mem_id] = rep_id
eliminated.append((mem_id, "h"))
elif (
pidentity >= options.min_identity_non_genes
and this_quality in options.quality_genes
and mem_quality not in options.quality_genes
):
if rep_coverage < mem_coverage - options.safety_coverage or rep_pid < mem_pid - options.safety_pide:
options.stdlog.write(
"# WARNING: not removing possibly good prediction: rep = %s, mem = %s, rep_cov=%i, rep_pid=%i, mem_cov=%i, mem_pid=%i\n"
% (rep_id, mem_id, rep_coverage, rep_pid, mem_coverage, mem_pid)
)
else:
eliminated_predictions[mem_id] = rep_id
eliminated.append((mem_id, "l"))
return eliminated
def PrintCluster(cluster,
cluster_id,
lengths,
peptide_sequences=None,
regex_preferred=None):
"""print a cluster.
Take longest sequence as representative. If preferred is given, only take
genes matching preferred identifier.
"""
if regex_preferred:
rx = re.compile(regex_preferred)
else:
rx = None
max_al = 0
max_pl = 0
rep_a = None
rep_p = None
for c in cluster:
l = 0
if c in lengths:
l = lengths[c]
if l > max_al:
max_al = l
rep_a = c
if rx and rx.search(c) and l > max_pl:
max_pl = l
rep_p = c
if max_pl > 0:
max_l = max_pl
rep = rep_p
else:
max_l = max_al
rep = rep_a
for mem in cluster:
l = 0
if mem in lengths:
l = lengths[mem]
if peptide_sequences:
map_rep2mem = alignlib_lite.makeAlignmentVector()
if rep == mem and rep in lengths:
alignlib_lite.addDiagonal2Alignment(
map_rep2mem, 1, lengths[rep], 0)
elif mem in peptide_sequences and \
rep in peptide_sequences:
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -1.0)
alignator.align(map_rep2mem,
alignlib_lite.makeSequence(
peptide_sequences[rep]),
alignlib_lite.makeSequence(peptide_sequences[mem]))
f = alignlib_lite.AlignmentFormatEmissions(map_rep2mem)
print string.join(map(str, (rep, mem, l, f)), "\t")
else:
print string.join(map(str, (rep, mem, l)), "\t")
sys.stdout.flush()
return cluster_id
def FilterConflicts(old_predictions, new_predictions, removed_predictions,
min_overlap, peptide_sequences):
"""remove conflicts.
Remove overlapping entries between different queries.
Only remove those sequences, which are alignable.
If they are alignable, take the sequence with the highest score and highest coverage.
(Take both, if score and coverage are not correlated.)
"""
##########################################################################
# sort predictions by genomic region
if isinstance(old_predictions, PredictionFile.PredictionFile):
old_predictions.sort(
('mSbjctToken', 'mSbjctStrand', 'mSbjctGenomeFrom', 'mSbjctGenomeTo'))
else:
old_predictions.sort(lambda x, y: cmp((x.mSbjctToken, x.mSbjctStrand, x.mSbjctGenomeFrom, x.mSbjctGenomeTo),
(y.mSbjctToken, y.mSbjctStrand, y.mSbjctGenomeFrom, y.mSbjctGenomeTo)))
##########################################################################
# filter predictions and resolve conflicts based on genomic overlap
# deleted segments are put in a temporary storage space.
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, param_gop, param_gep)
result = alignlib_lite.makeAlignmentVector()
alignments = {}
noverlaps = 0
nredundants = 0
nnew = 0
last_prediction = None
for this_prediction in old_predictions:
try:
(this_query_peptide, this_query_status, this_query_gene,
this_query_transcript) = \
re.split("\s+", this_prediction.mQueryToken)
except ValueError:
this_query_gene = None
if not last_prediction:
last_prediction = this_prediction
last_query_gene = this_query_gene
continue
overlap = min(last_prediction.mSbjctGenomeTo,
this_prediction.mSbjctGenomeTo) -\
max(last_prediction.mSbjctGenomeFrom,
this_prediction.mSbjctGenomeFrom)
union = max(last_prediction.mSbjctGenomeTo,
this_prediction.mSbjctGenomeTo) -\
min(last_prediction.mSbjctGenomeFrom,
this_prediction.mSbjctGenomeFrom)
# resolve overlap between different genes
if overlap > 0 and \
(last_query_gene != this_query_gene or
last_query_gene is None):
noverlaps += 1
relative_overlap = 100 * overlap / union
# Start conflict resolution, if overlap is above threshold.
# Keep higher scoring segment.
#
# Check if queries are homologous.
if relative_overlap >= param_max_percent_overlap:
if peptide_sequences:
if last_prediction.mQueryToken < this_prediction.mQueryToken:
key = "%s-%s" % (last_prediction.mQueryToken,
this_prediction.mQueryToken)
else:
key = "%s-%s" % (this_prediction.mQueryToken,
last_prediction.mQueryToken)
if not alignments.has_key(key):
result.clear()
alignator.align(result,
alignlib_lite.makeSequence(
peptide_sequences[this_prediction.mQueryToken]),
alignlib_lite.makeSequence(peptide_sequences[last_prediction.mQueryToken]))
alignments[key] = result.getScore()
if result.getScore() >= param_min_score_overlap:
nredundants += 1
if alignments[key] >= param_min_score_overlap:
is_overlap = 1
else:
is_overlap = 0
else:
is_overlap = 1
else:
is_overlap = 0
else:
is_overlap = 0
if is_overlap:
# take best prediction. If difference is very small, set
# difference to 0 (difference does not matter). In this case,
# the first prediction is taken.
d1 = last_prediction.mQueryCoverage - \
this_prediction.mQueryCoverage
if float(abs(d1)) / float(last_prediction.mQueryCoverage) < param_conflicts_min_difference:
d1 = 0
d2 = last_prediction.score - this_prediction.score
if float(abs(d2)) / float(this_prediction.score) < param_conflicts_min_difference:
d2 = 0
if d1 >= 0 and d2 >= 0:
if param_loglevel >= 2:
print "# CONFLICT: kept %i(%s-%i), overlap=%i(%5.2f), removed: %s" % (last_prediction.mPredictionId,
last_prediction.mQueryToken,
last_prediction.mSbjctGenomeFrom,
overlap, relative_overlap,
str(this_prediction))
if param_benchmarks:
if CheckBenchmark(this_prediction, last_prediction):
print "# BENCHMARK KEPT with overlap=%i(%5.2f): %s" % (overlap, relative_overlap,
str(last_prediction))
removed_predictions.append(this_prediction)
continue
elif d1 <= 0 and d2 <= 0:
if param_loglevel >= 2:
print "# CONFLICT: kept %i(%s-%i), overlap=%i(%5.2f), removed: %s" % (this_prediction.mPredictionId,
this_prediction.mQueryToken,
this_prediction.mSbjctGenomeFrom,
overlap, relative_overlap,
str(last_prediction))
if param_benchmarks:
if CheckBenchmark(last_prediction, this_prediction):
print "# BENCHMARK KEPT with overlap=%i(%5.2f): %s" % (overlap, relative_overlap,
str(this_prediction))
removed_predictions.append(last_prediction)
last_prediction = this_prediction
last_query_gene = this_query_gene
continue
else:
if param_loglevel >= 2:
print "# CONFLICT: non-correlated score/coverage. Keeping both %i(%s-%i) (%5.2f/%i/%i) and %i(%s-%i) (%5.2f/%i/%i)" % \
(this_prediction.mPredictionId,
this_prediction.mQueryToken, this_prediction.mSbjctGenomeFrom,
this_prediction.score, this_prediction.mQueryCoverage,
this_prediction.mPercentIdentity,
last_prediction.mPredictionId,
last_prediction.mQueryToken, last_prediction.mSbjctGenomeFrom,
last_prediction.score, last_prediction.mQueryCoverage,
last_prediction.mPercentIdentity)
new_predictions.append(last_prediction)
nnew += 1
last_query_gene = this_query_gene
last_prediction = this_prediction
new_predictions.append(last_prediction)
nnew += 1
if param_loglevel >= 1:
print "# calculated %i alignments for %i potential conflicts (%i above threshold)" % \
(len(alignments), noverlaps, nredundants)
return nnew
def GetOrthologTranscripts(transcripts1, peptides1, cds1, transcripts2,
peptides2, cds2):
"""sort out ortholog relationships between
transcripts of orthologous genes.
Orthologs have:
the same number of exons
compatible intron/exon boundaries
For the remaining transcript pairs, take reciprocal bet hits.
I see the following:
0: 0(100%), 1: 0(94%), 2: 0,1(100%)
0: 0(100%), 1: 0,1,2(100%)
Selecting 1-0 first, would result in a suboptimal match, because one transcript
is longer than the other, while matching up 0-0 and 2-1 would be better.
Objective function: it is the maximal matching/assignment problem. Use greedy
implementation instead. Assign as much as possible according to descending weights.
"""
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -2.0)
# for long sequence: use dot alignment with tuple size of three
dottor = alignlib_lite.makeAlignatorTuples(3)
alignator_dots = alignlib_lite.makeAlignatorDotsSquared(
param_gop, param_gep, dottor)
seqs1 = map(lambda x: alignlib_lite.makeSequence(peptides1[x[0]]),
transcripts1)
seqs2 = map(lambda x: alignlib_lite.makeSequence(peptides2[x[0]]),
transcripts2)
if param_loglevel >= 4:
print "# building sequence 1"
for i in range(len(seqs1)):
if not cds1.has_key(transcripts1[i][0]):
if param_loglevel >= 4:
print "# %s not found" % transcripts1[i][0]
if param_loglevel >= 4:
print "# building sequence 2"
for i in range(len(seqs2)):
if not cds2.has_key(transcripts2[i][0]):
if param_loglevel >= 4:
print "# %s not found" % transcripts1[i][0]
if param_loglevel >= 4:
print "# all-vs-all alignment"
# do all versus all alignment
alis1 = []
alis2 = []
for i in range(len(seqs1)):
alis1.append([])
for i in range(len(seqs2)):
alis2.append([])
if param_loglevel >= 3:
print "#################################"
for i in range(len(seqs1)):
for cd in cds1[transcripts1[i][0]]:
print "#", str(cd)
print "# versus"
for i in range(len(seqs2)):
for cd in cds2[transcripts2[i][0]]:
print "#", str(cd)
sys.stdout.flush()
weights = {}
for i in range(len(seqs1)):
prediction_id1, sbjct_token1, sbjct_strand1, sbjct_from1, sbjct_to1 = transcripts1[
i]
for j in range(len(seqs2)):
prediction_id2, sbjct_token2, sbjct_strand2, sbjct_from2, sbjct_to2 = transcripts2[
j]
map_a2b = alignlib_lite.makeAlignmentVector()
m = seqs1[i].getLength() * seqs2[j].getLength()
if param_loglevel >= 3:
print "# Starting alignment of pair (%i,%i) of lengths %s:%i and %s:%i" %\
(i, j, prediction_id1, seqs1[
i].getLength(), prediction_id2, seqs2[j].getLength())
sys.stdout.flush()
if m > param_max_matrix_size:
# switch to tuple alignment if sequences are too large
if param_loglevel >= 2:
print "# WARNING: sequences are of length %i and %i: switching to dot alignment." % (
seqs1[i].getLength(), seqs2[j].getLength())
sys.stdout.flush()
alignator_dots.align(map_a2b, seqs1[i], seqs2[j])
else:
alignator.align(map_a2b, seqs1[i], seqs2[j])
coverage_a = 100.0 * \
(map_a2b.getRowTo() - map_a2b.getRowFrom() + 1) / \
seqs1[i].getLength()
coverage_b = 100.0 * \
(map_a2b.getColTo() - map_a2b.getColFrom() + 1) / \
seqs2[j].getLength()
# get copy of cds, but only those overlapping with alignment
c1 = Exons.GetExonsRange(
cds1[prediction_id1], (map_a2b.getRowFrom() - 1) * 3,
(map_a2b.getRowTo()) * 3 + 1,
full=False,
min_overlap=param_min_alignment_exon_overlap,
min_exon_size=param_min_exon_size)
c2 = Exons.GetExonsRange(
cds2[prediction_id2], (map_a2b.getColFrom() - 1) * 3,
(map_a2b.getColTo()) * 3 + 1,
full=False,
min_overlap=param_min_alignment_exon_overlap,
min_exon_size=param_min_exon_size)
# check exon boundaries, look at starts, skip first exon
def MyMap(a, x):
while x <= a.getRowTo():
c = a.mapRowToCol(x)
if c:
return c
x += 1
else:
return 0
mapped_boundaries = map(
lambda x: MyMap(map_a2b, x.mPeptideFrom / 3 + 1), c1[1:])
mapped_boundaries.sort()
reference_boundaries = map(lambda x: x.mPeptideFrom / 3 + 1,
c2[1:])
reference_boundaries.sort()
nmissed_cmp2ref = Exons.CountMissedBoundaries(
mapped_boundaries, reference_boundaries,
param_boundaries_max_slippage)
nmissed_ref2cmp = Exons.CountMissedBoundaries(
reference_boundaries, mapped_boundaries,
param_boundaries_max_slippage)
min_nmissed = min(nmissed_cmp2ref, nmissed_ref2cmp)
# set is_ok for the whole thing
# no intron: is ok
is_ok = 0
if (len(c1) == 1 and len(c2) == 1):
is_ok = 1
else:
# allow for missed boundaries, if param_boundaries_allow_missed
# > 0
if min_nmissed == 0:
is_ok = 1
else:
if param_boundaries_allow_missed and \
len(mapped_boundaries) >= param_boundaries_allow_missed and \
min_nmissed <= param_boundaries_max_missed:
is_ok = 1
cc = min(coverage_a, coverage_b)
if cc >= param_min_coverage:
is_ok_coverage = 1
else:
is_ok_coverage = 0
# check for missing introns
is_ok_exons = 1
if abs(len(c1) - len(c2)) != 0:
if param_missing_max_missing:
if ((abs(len(c1) - len(c2)) > param_missing_max_missing) or
(min(len(c1), len(c2)) < param_missing_min_present)):
is_ok_exons = 0
else:
is_ok_exons = 0
if param_loglevel >= 3:
print "# i=", i, "li=", len(c1), "j=", j, "lj=", len(c2), \
"boundaries_ok=", is_ok, \
"nexons_ok=", is_ok_exons, \
"missed_c2r=", nmissed_cmp2ref, \
"missed_r2c=", nmissed_ref2cmp, \
"min_cov=", cc, \
"mapped=", mapped_boundaries, \
"reference=", reference_boundaries
print "#", string.join(
map(str, (alignlib_lite.AlignmentFormatEmissions(map_a2b),
map_a2b.getNumGaps(), coverage_a, coverage_b)),
"\t")
sys.stdout.flush()
# dump out pairs
for method in param_write_pairs:
if method == "all":
print string.join(
map(str,
("pair", method, prediction_id1, prediction_id2,
sbjct_token1, sbjct_strand1, sbjct_from1,
sbjct_to1, seqs1[i].getLength(), sbjct_token2,
sbjct_strand2, sbjct_from2, sbjct_to2,
seqs2[j].getLength(), map_a2b.getRowFrom(),
map_a2b.getRowTo(), row_ali, map_a2b.getColFrom(),
map_a2b.getColTo(), col_ali, map_a2b.getNumGaps(),
coverage_a, coverage_b, nmissed_cmp2ref,
mapped_boundaries, nmissed_ref2cmp,
reference_boundaries, i, j, len(c1), len(c2), cc,
is_ok, is_ok_exons, is_ok_coverage)), "\t")
elif method == "alignment":
print string.join(
map(str,
("pair", method, prediction_id1, prediction_id2,
map_a2b.getRowFrom(), map_a2b.getRowTo(), row_ali,
map_a2b.getColFrom(), map_a2b.getColTo(), col_ali,
map_a2b.getNumGaps(), coverage_a, coverage_b)),
"\t")
elif method == "location":
print string.join(
map(str, ("pair", method, prediction_id1,
prediction_id2, sbjct_token1, sbjct_strand1,
sbjct_from1, sbjct_to1, seqs1[i].getLength(),
sbjct_token2, sbjct_strand2, sbjct_from2,
sbjct_to2, seqs2[j].getLength())), "\t")
if not is_ok_exons:
if param_loglevel >= 4:
print "# rejected %i and %i: too many exons difference." % (
i, j)
continue
if param_check_exon_boundaries:
if not is_ok:
continue
if cc < param_min_coverage:
continue
if not weights.has_key(cc):
weights[cc] = []
alis1[i].append((coverage_a, j))
alis2[j].append((coverage_b, i))
weights[cc].append((i, j, map_a2b))
# sort out alignments
ww = weights.keys()
ww.sort()
ww.reverse()
pairs = []
assigned1 = {}
assigned2 = {}
if param_loglevel >= 3:
print "# alis1=", alis1
print "# alis2=", alis2
print "# --------------------------------------"
for w in ww:
for i, j, map_a2b in weights[w]:
if not assigned1.has_key(i) and not assigned2.has_key(j):
pairs.append((transcripts1[i], transcripts2[j], w, map_a2b))
assigned1[i] = 1
assigned2[j] = 1
if len(assigned1) == len(transcripts1):
break
if len(assigned2) == len(transcripts2):
break
return pairs
elif param_mode == "transcripts":
transcript_ids1 = {}
transcript_ids2 = {}
for x, y in pairs:
transcript_ids1[x] = 1
transcript_ids2[y] = 1
transcripts1, transcripts2, cds1, cds2 = ReadTranscriptsAndCds(
transcript_ids1, transcript_ids2)
if param_loglevel >= 1:
print "# reading has finished."
sys.stdout.flush()
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -2.0)
for q1, q2 in pairs:
ninput += 1
if param_loglevel >= 1:
print "# processing %s and %s" % (q1, q2)
if q1 in transcripts1 and q2 in transcripts2:
map_a2b = alignlib_lite.makeAlignmentVector()
alignator.align(map_a2b,
alignlib_lite.makeSequence(peptides1[q1]),
alignlib_lite.makeSequence(peptides2[q2]))
nintrons,
nsplits,
nstopcodons,
pidentity,
psimilarity,
sequence,
sbjct_genome_from,
sbjct_genome_to,
map_query2genome
FROM %s AS p
WHERE p.sbjct_token = '%s' AND
p.sbjct_strand = '%s' AND
OVERLAP( %i, %i, p.sbjct_genome_from, sbjct_genome_to) > 0
"""
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, param_gop, param_gep)
map_reference2target = alignlib_lite.makeAlignmentVector()
assignment_id = 0
for line in cr.fetchall():
reference = PredictionParser.PredictionParserEntry()
reference.FillFromTable(line)
ct = dbhandle.cursor()
ct.execute(statement % (param_tablename_predictions_target,
reference.mSbjctToken, reference.mSbjctStrand,
reference.mSbjctGenomeFrom, reference.mSbjctGenomeTo))
reference_exons = Exons.Alignment2Exons(reference.mMapPeptide2Genome,
0,
def PrintCluster(cluster,
cluster_id,
lengths,
peptide_sequences=None,
regex_preferred=None):
"""print a cluster.
Take longest sequence as representative. If preferred is given, only take
genes matching preferred identifier.
"""
if regex_preferred:
rx = re.compile(regex_preferred)
else:
rx = None
max_al = 0
max_pl = 0
rep_a = None
rep_p = None
for c in cluster:
l = 0
if c in lengths: l = lengths[c]
if l > max_al:
max_al = l
rep_a = c
if rx and rx.search(c) and l > max_pl:
max_pl = l
rep_p = c
if max_pl > 0:
max_l = max_pl
rep = rep_p
else:
max_l = max_al
rep = rep_a
for mem in cluster:
l = 0
if mem in lengths: l = lengths[mem]
if peptide_sequences:
map_rep2mem = alignlib_lite.makeAlignmentVector()
if rep == mem and rep in lengths:
alignlib_lite.addDiagonal2Alignment(map_rep2mem, 1,
lengths[rep], 0)
elif mem in peptide_sequences and \
rep in peptide_sequences:
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -1.0)
alignator.align(
map_rep2mem,
alignlib_lite.makeSequence(peptide_sequences[rep]),
alignlib_lite.makeSequence(peptide_sequences[mem]))
f = alignlib_lite.AlignmentFormatEmissions(map_rep2mem)
print string.join(map(str, (rep, mem, l, f)), "\t")
else:
print string.join(map(str, (rep, mem, l)), "\t")
sys.stdout.flush()
return cluster_id
def GetOrthologTranscripts(transcripts1, peptides1, cds1,
transcripts2, peptides2, cds2):
"""sort out ortholog relationships between
transcripts of orthologous genes.
Orthologs have:
the same number of exons
compatible intron/exon boundaries
For the remaining transcript pairs, take reciprocal bet hits.
I see the following:
0: 0(100%), 1: 0(94%), 2: 0,1(100%)
0: 0(100%), 1: 0,1,2(100%)
Selecting 1-0 first, would result in a suboptimal match, because one transcript
is longer than the other, while matching up 0-0 and 2-1 would be better.
Objective function: it is the maximal matching/assignment problem. Use greedy
implementation instead. Assign as much as possible according to descending weights.
"""
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -2.0)
# for long sequence: use dot alignment with tuple size of three
dottor = alignlib_lite.makeAlignatorTuples(3)
alignator_dots = alignlib_lite.makeAlignatorDotsSquared(
param_gop, param_gep, dottor)
seqs1 = map(lambda x: alignlib_lite.makeSequence(
peptides1[x[0]]), transcripts1)
seqs2 = map(lambda x: alignlib_lite.makeSequence(
peptides2[x[0]]), transcripts2)
if param_loglevel >= 4:
print "# building sequence 1"
for i in range(len(seqs1)):
if not cds1.has_key(transcripts1[i][0]):
if param_loglevel >= 4:
print "# %s not found" % transcripts1[i][0]
if param_loglevel >= 4:
print "# building sequence 2"
for i in range(len(seqs2)):
if not cds2.has_key(transcripts2[i][0]):
if param_loglevel >= 4:
print "# %s not found" % transcripts1[i][0]
if param_loglevel >= 4:
print "# all-vs-all alignment"
# do all versus all alignment
alis1 = []
alis2 = []
for i in range(len(seqs1)):
alis1.append([])
for i in range(len(seqs2)):
alis2.append([])
if param_loglevel >= 3:
print "#################################"
for i in range(len(seqs1)):
for cd in cds1[transcripts1[i][0]]:
print "#", str(cd)
print "# versus"
for i in range(len(seqs2)):
for cd in cds2[transcripts2[i][0]]:
print "#", str(cd)
sys.stdout.flush()
weights = {}
for i in range(len(seqs1)):
prediction_id1, sbjct_token1, sbjct_strand1, sbjct_from1, sbjct_to1 = transcripts1[
i]
for j in range(len(seqs2)):
prediction_id2, sbjct_token2, sbjct_strand2, sbjct_from2, sbjct_to2 = transcripts2[
j]
map_a2b = alignlib_lite.makeAlignmentVector()
m = seqs1[i].getLength() * seqs2[j].getLength()
if param_loglevel >= 3:
print "# Starting alignment of pair (%i,%i) of lengths %s:%i and %s:%i" %\
(i, j, prediction_id1, seqs1[
i].getLength(), prediction_id2, seqs2[j].getLength())
sys.stdout.flush()
if m > param_max_matrix_size:
# switch to tuple alignment if sequences are too large
if param_loglevel >= 2:
print "# WARNING: sequences are of length %i and %i: switching to dot alignment." % (seqs1[i].getLength(), seqs2[j].getLength())
sys.stdout.flush()
alignator_dots.align(map_a2b, seqs1[i], seqs2[j])
else:
alignator.align(map_a2b, seqs1[i], seqs2[j])
coverage_a = 100.0 * \
(map_a2b.getRowTo() - map_a2b.getRowFrom() + 1) / \
seqs1[i].getLength()
coverage_b = 100.0 * \
(map_a2b.getColTo() - map_a2b.getColFrom() + 1) / \
seqs2[j].getLength()
# get copy of cds, but only those overlapping with alignment
c1 = Exons.GetExonsRange(cds1[prediction_id1],
(map_a2b.getRowFrom() - 1) * 3,
(map_a2b.getRowTo()) * 3 + 1,
full=False,
min_overlap=param_min_alignment_exon_overlap,
min_exon_size=param_min_exon_size)
c2 = Exons.GetExonsRange(cds2[prediction_id2],
(map_a2b.getColFrom() - 1) * 3,
(map_a2b.getColTo()) * 3 + 1,
full=False,
min_overlap=param_min_alignment_exon_overlap,
min_exon_size=param_min_exon_size)
# check exon boundaries, look at starts, skip first exon
def MyMap(a, x):
while x <= a.getRowTo():
c = a.mapRowToCol(x)
if c:
return c
x += 1
else:
return 0
mapped_boundaries = map(
lambda x: MyMap(map_a2b, x.mPeptideFrom / 3 + 1), c1[1:])
mapped_boundaries.sort()
reference_boundaries = map(
lambda x: x.mPeptideFrom / 3 + 1, c2[1:])
reference_boundaries.sort()
nmissed_cmp2ref = Exons.CountMissedBoundaries(
mapped_boundaries, reference_boundaries, param_boundaries_max_slippage)
nmissed_ref2cmp = Exons.CountMissedBoundaries(
reference_boundaries, mapped_boundaries, param_boundaries_max_slippage)
min_nmissed = min(nmissed_cmp2ref, nmissed_ref2cmp)
# set is_ok for the whole thing
# no intron: is ok
is_ok = 0
if (len(c1) == 1 and len(c2) == 1):
is_ok = 1
else:
# allow for missed boundaries, if param_boundaries_allow_missed
# > 0
if min_nmissed == 0:
is_ok = 1
else:
if param_boundaries_allow_missed and \
len(mapped_boundaries) >= param_boundaries_allow_missed and \
min_nmissed <= param_boundaries_max_missed:
is_ok = 1
cc = min(coverage_a, coverage_b)
if cc >= param_min_coverage:
is_ok_coverage = 1
else:
is_ok_coverage = 0
# check for missing introns
is_ok_exons = 1
if abs(len(c1) - len(c2)) != 0:
if param_missing_max_missing:
if ((abs(len(c1) - len(c2)) > param_missing_max_missing) or
(min(len(c1), len(c2)) < param_missing_min_present)):
is_ok_exons = 0
else:
is_ok_exons = 0
if param_loglevel >= 3:
print "# i=", i, "li=", len(c1), "j=", j, "lj=", len(c2), \
"boundaries_ok=", is_ok, \
"nexons_ok=", is_ok_exons, \
"missed_c2r=", nmissed_cmp2ref, \
"missed_r2c=", nmissed_ref2cmp, \
"min_cov=", cc, \
"mapped=", mapped_boundaries, \
"reference=", reference_boundaries
print "#", string.join(map(str, (alignlib_lite.AlignmentFormatEmissions(map_a2b),
map_a2b.getNumGaps(), coverage_a, coverage_b)), "\t")
sys.stdout.flush()
# dump out pairs
for method in param_write_pairs:
if method == "all":
print string.join(map(str, (
"pair", method,
prediction_id1,
prediction_id2,
sbjct_token1, sbjct_strand1, sbjct_from1, sbjct_to1, seqs1[
i].getLength(),
sbjct_token2, sbjct_strand2, sbjct_from2, sbjct_to2, seqs2[
j].getLength(),
map_a2b.getRowFrom(), map_a2b.getRowTo(), row_ali,
map_a2b.getColFrom(), map_a2b.getColTo(), col_ali,
map_a2b.getNumGaps(), coverage_a, coverage_b,
nmissed_cmp2ref, mapped_boundaries,
nmissed_ref2cmp, reference_boundaries,
i, j, len(c1), len(c2), cc, is_ok, is_ok_exons, is_ok_coverage)), "\t")
elif method == "alignment":
print string.join(map(str, (
"pair", method,
prediction_id1, prediction_id2,
map_a2b.getRowFrom(), map_a2b.getRowTo(), row_ali,
map_a2b.getColFrom(), map_a2b.getColTo(), col_ali,
map_a2b.getNumGaps(), coverage_a, coverage_b)), "\t")
elif method == "location":
print string.join(map(str, (
"pair", method,
prediction_id1,
prediction_id2,
sbjct_token1, sbjct_strand1, sbjct_from1, sbjct_to1, seqs1[
i].getLength(),
sbjct_token2, sbjct_strand2, sbjct_from2, sbjct_to2, seqs2[j].getLength())), "\t")
if not is_ok_exons:
if param_loglevel >= 4:
print "# rejected %i and %i: too many exons difference." % (i, j)
continue
if param_check_exon_boundaries:
if not is_ok:
continue
if cc < param_min_coverage:
continue
if not weights.has_key(cc):
weights[cc] = []
alis1[i].append((coverage_a, j))
alis2[j].append((coverage_b, i))
weights[cc].append((i, j, map_a2b))
# sort out alignments
ww = weights.keys()
ww.sort()
ww.reverse()
pairs = []
assigned1 = {}
assigned2 = {}
if param_loglevel >= 3:
print "# alis1=", alis1
print "# alis2=", alis2
print "# --------------------------------------"
for w in ww:
for i, j, map_a2b in weights[w]:
if not assigned1.has_key(i) and not assigned2.has_key(j):
pairs.append((transcripts1[i], transcripts2[j], w, map_a2b))
assigned1[i] = 1
assigned2[j] = 1
if len(assigned1) == len(transcripts1):
break
if len(assigned2) == len(transcripts2):
break
return pairs
elif param_mode == "transcripts":
transcript_ids1 = {}
transcript_ids2 = {}
for x, y in pairs:
transcript_ids1[x] = 1
transcript_ids2[y] = 1
transcripts1, transcripts2, cds1, cds2 = ReadTranscriptsAndCds(
transcript_ids1, transcript_ids2)
if param_loglevel >= 1:
print "# reading has finished."
sys.stdout.flush()
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, -10.0, -2.0)
for q1, q2 in pairs:
ninput += 1
if param_loglevel >= 1:
print "# processing %s and %s" % (q1, q2)
if q1 in transcripts1 and q2 in transcripts2:
map_a2b = alignlib_lite.makeAlignmentVector()
alignator.align(map_a2b,
alignlib_lite.makeSequence(peptides1[q1]),
alignlib_lite.makeSequence(peptides2[q2]))
def ProcessRegion(predictions, region_id, region,
peptide_sequences=None,
filter_queries={}):
"""process a set of matches to a region.
resolve region according to homology.
"""
if options.loglevel >= 3:
options.stdlog.write(
"###################################################################\n")
options.stdlog.write(
"# resolving %i predictions in region %s\n" % (len(predictions), str(region)))
sys.stdout.flush()
predictions.sort(lambda x, y: cmp(x.score, y.score))
predictions.reverse()
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, options.gop, options.gep)
result = alignlib_lite.makeAlignmentVector()
cluster = []
map_sequence2cluster = range(0, len(predictions))
edges = []
noutput, nskipped = 0, 0
if peptide_sequences:
for x in range(len(predictions)):
if options.loglevel >= 5:
options.stdlog.write("# filtering from %i with prediction %i: %s\n" % (
x, predictions[x].mPredictionId, predictions[x].mQueryToken))
sys.stdout.flush()
if map_sequence2cluster[x] != x:
continue
region_id += 1
edges = []
if predictions[x].mQueryToken not in filter_queries:
edges.append(predictions[x])
else:
nskipped += 1
for y in range(x + 1, len(predictions)):
if map_sequence2cluster[y] != y:
continue
if predictions[x].mQueryToken < predictions[y].mQueryToken:
key = "%s-%s" % (predictions[x].mQueryToken,
predictions[y].mQueryToken)
else:
key = "%s-%s" % (predictions[y].mQueryToken,
predictions[x].mQueryToken)
# check if predictions are overlapping on the genomic sequence
if min(predictions[x].mSbjctGenomeTo, predictions[y].mSbjctGenomeTo) - \
max(predictions[x].mSbjctGenomeFrom, predictions[y].mSbjctGenomeFrom) < 0:
if options.loglevel >= 4:
options.stdlog.write("# alignment of predictions %i and %i: no overlap on genomic sequence, thus skipped\n" %
(predictions[x].mPredictionId,
predictions[y].mPredictionId))
sys.stdout.flush()
continue
if not global_alignments.has_key(key):
seq1 = peptide_sequences[predictions[x].mQueryToken]
seq2 = peptide_sequences[predictions[y].mQueryToken]
result.clear()
s1 = alignlib_lite.makeSequence(seq1)
s2 = alignlib_lite.makeSequence(seq2)
alignator.align(result, s1, s2)
c1 = 100 * \
(result.getRowTo() - result.getRowFrom()) / len(seq1)
c2 = 100 * \
(result.getColTo() - result.getColFrom()) / len(seq2)
min_cov = min(c1, c2)
max_cov = max(c1, c2)
identity = alignlib_lite.calculatePercentIdentity(
result, s1, s2) * 100
# check if predictions overlap and they are homologous
if result.getScore() >= options.overlap_min_score and \
max_cov >= options.overlap_max_coverage and \
min_cov >= options.overlap_min_coverage and \
identity >= options.overlap_min_identity:
global_alignments[key] = True
else:
global_alignments[key] = False
if options.loglevel >= 4:
options.stdlog.write("# alignment=%s score=%i pid=%5.2f c1=%i c2=%i min_cov=%i max_cov=%i homolog=%s\n" %
(key,
result.getScore(),
identity,
c1, c2, min_cov, max_cov,
global_alignments[key]))
sys.stdout.flush()
if global_alignments[key]:
map_sequence2cluster[y] = x
if predictions[y].mQueryToken not in filter_queries:
edges.append(predictions[y])
else:
nskipped += 1
noutput += PrintEdges(region_id, region, edges)
return region_id, noutput, nskipped
nintrons,
nsplits,
nstopcodons,
pidentity,
psimilarity,
sequence,
sbjct_genome_from,
sbjct_genome_to,
map_query2genome
FROM %s AS p
WHERE p.sbjct_token = '%s' AND
p.sbjct_strand = '%s' AND
OVERLAP( %i, %i, p.sbjct_genome_from, sbjct_genome_to) > 0
"""
alignator = alignlib_lite.makeAlignatorDPFull(
alignlib_lite.ALIGNMENT_LOCAL, param_gop, param_gep)
map_reference2target = alignlib_lite.makeAlignmentVector()
assignment_id = 0
for line in cr.fetchall():
reference = PredictionParser.PredictionParserEntry()
reference.FillFromTable(line)
ct = dbhandle.cursor()
ct.execute(statement %
(param_tablename_predictions_target, reference.mSbjctToken,
reference.mSbjctStrand, reference.mSbjctGenomeFrom,
reference.mSbjctGenomeTo))
reference_exons = Exons.Alignment2Exons(reference.mMapPeptide2Genome,
def EliminateRedundantEntries( rep,
data,
eliminated_predictions,
options,
peptides,
extended_peptides,
filter_quality = None,
this_quality = None ):
"""eliminate redundant entries in a set."""
eliminated = []
rep_id = rep.transcript_id
rep_coverage, rep_pid = rep.mQueryCoverage, rep.mPid
alignator = alignlib_lite.makeAlignatorDPFull( alignlib_lite.ALIGNMENT_LOCAL, options.gop, options.gep )
result = alignlib_lite.makeAlignmentVector()
rep_seq = peptides[rep_id]
rep_extended_seq = extended_peptides[rep_id]
for entry in data:
mem_id, mem_coverage, mem_pid, mem_quality = ( entry.transcript_id,
entry.mQueryCoverage,
entry.mPid,
entry.mQuality )
mem_seq = peptides[mem_id]
mem_extended_seq = extended_peptides[mem_id]
if options.loglevel >= 4:
options.stdlog.write( "# processing: id=%s class=%s\n" % (mem_id, mem_quality))
if mem_id in eliminated_predictions: continue
if mem_extended_seq == rep_extended_seq:
eliminated_predictions[mem_id] = rep_id
eliminated.append( (mem_id, "i") )
elif mem_extended_seq in rep_extended_seq:
eliminated_predictions[mem_id] = rep_id
eliminated.append( (mem_id, "p") )
else:
if mem_quality != this_quality or \
mem_quality in options.quality_exclude_same:
seq1 = alignlib_lite.makeSequence( str(rep_seq) )
seq2 = alignlib_lite.makeSequence( str(mem_seq) )
alignator.align( result, seq1, seq2 )
if options.loglevel >= 5:
options.stdlog.write( "# ali\n%s\n" % alignlib_lite.AlignmentFormatExplicit( result, seq1, seq2 ) )
pidentity = 100 * alignlib_lite.calculatePercentIdentity( result, seq1, seq2 )
num_gaps = result.getNumGaps()
if options.loglevel >= 4:
options.stdlog.write( "# processing: id=%s class=%s pid=%5.2f rep_cov=%i mem_cov=%i\n" %\
( mem_id, mem_quality, pidentity, rep_coverage, mem_coverage ) )
if pidentity >= options.min_identity:
keep = False
if rep_coverage < mem_coverage - options.safety_coverage or \
rep_pid < mem_pid - options.safety_pide:
keep = True
reason = "covpid"
elif num_gaps >= options.max_gaps and \
mem_coverage > rep_coverage - options.safety_coverage:
keep = True
reason = "gaps"
elif mem_coverage >= rep_coverage - options.safety_coverage and \
100 * (result.getColTo() - result.getColFrom()) / len(mem_seq) < options.max_member_coverage:
keep = True
reason = "memcov"
if keep:
options.stdlog.write( "# WARNING: not removing possibly good prediction: %s: rep = %s, mem = %s, rep_cov=%i, rep_pid=%i, mem_cov=%i, mem_pid=%i\n" %\
(reason, rep_id, mem_id, rep_coverage, rep_pid, mem_coverage, mem_pid) )
else:
eliminated_predictions[mem_id] = rep_id
eliminated.append( (mem_id, "h") )
elif pidentity >= options.min_identity_non_genes and \
this_quality in options.quality_genes and \
mem_quality not in options.quality_genes:
if rep_coverage < mem_coverage - options.safety_coverage or \
rep_pid < mem_pid - options.safety_pide:
options.stdlog.write( "# WARNING: not removing possibly good prediction: rep = %s, mem = %s, rep_cov=%i, rep_pid=%i, mem_cov=%i, mem_pid=%i\n" %\
(rep_id, mem_id, rep_coverage, rep_pid, mem_coverage, mem_pid) )
else:
eliminated_predictions[mem_id] = rep_id
eliminated.append( (mem_id, "l") )
return eliminated