def testParse(self):
f = open("./SCOP/dir.cla.scop.txt_test")
try:
cla = f.read()
f.close()
f = open("./SCOP/dir.des.scop.txt_test")
des = f.read()
f.close()
f = open("./SCOP/dir.hie.scop.txt_test")
hie = f.read()
finally:
f.close()
scop = Scop(StringIO(cla), StringIO(des), StringIO(hie))
cla_out = StringIO()
scop.write_cla(cla_out)
lines = zip(cla.rstrip().split("\n"),
cla_out.getvalue().rstrip().split("\n"))
for expected_line, line in lines:
self.assertTrue(self._compare_cla_lines(expected_line, line))
des_out = StringIO()
scop.write_des(des_out)
self.assertEqual(des_out.getvalue(), des)
hie_out = StringIO()
scop.write_hie(hie_out)
self.assertEqual(hie_out.getvalue(), hie)
domain = scop.getDomainBySid("d1hbia_")
self.assertEqual(domain.sunid, 14996)
domains = scop.getDomains()
self.assertEqual(len(domains), 14)
self.assertEqual(domains[4].sunid, 14988)
dom = scop.getNodeBySunid(-111)
self.assertEqual(dom, None)
dom = scop.getDomainBySid("no such domain")
self.assertEqual(dom, None)
def test_get_descendents(self):
"""Test getDescendents method."""
scop = Scop(dir_path="SCOP", version="test")
fold = scop.getNodeBySunid(46457)
# get px descendents
domains = fold.getDescendents("px")
self.assertEqual(len(domains), 14)
for d in domains:
self.assertEqual(d.type, "px")
sfs = fold.getDescendents("superfamily")
self.assertEqual(len(sfs), 1)
for d in sfs:
self.assertEqual(d.type, "sf")
# cl has no cl descendent
cl = fold.getDescendents("cl")
self.assertEqual(cl, [])
def testParse(self):
f = open("./SCOP/dir.cla.scop.txt_test")
try:
cla = f.read()
f.close()
f = open("./SCOP/dir.des.scop.txt_test")
des = f.read()
f.close()
f = open("./SCOP/dir.hie.scop.txt_test")
hie = f.read()
finally:
f.close()
scop = Scop(StringIO(cla), StringIO(des), StringIO(hie))
cla_out = StringIO()
scop.write_cla(cla_out)
lines = zip(cla.rstrip().split('\n'),
cla_out.getvalue().rstrip().split('\n'))
for expected_line, line in lines:
self.assertTrue(self._compare_cla_lines(expected_line, line))
des_out = StringIO()
scop.write_des(des_out)
self.assertEqual(des_out.getvalue(), des)
hie_out = StringIO()
scop.write_hie(hie_out)
self.assertEqual(hie_out.getvalue(), hie)
domain = scop.getDomainBySid("d1hbia_")
self.assertEqual(domain.sunid, 14996)
domains = scop.getDomains()
self.assertEqual(len(domains), 14)
self.assertEqual(domains[4].sunid, 14988)
dom = scop.getNodeBySunid(-111)
self.assertEqual(dom, None)
dom = scop.getDomainBySid("no such domain")
self.assertEqual(dom, None)
def test_get_descendents(self):
"""Test getDescendents method"""
scop = Scop(dir_path="SCOP", version="test")
fold = scop.getNodeBySunid(46457)
# get px descendents
domains = fold.getDescendents('px')
self.assertEqual(len(domains), 14)
for d in domains:
self.assertEqual(d.type, 'px')
sfs = fold.getDescendents('superfamily')
self.assertEqual(len(sfs), 1)
for d in sfs:
self.assertEqual(d.type, 'sf')
# cl has no cl descendent
cl = fold.getDescendents('cl')
self.assertEqual(cl, [])
def blast(algo):
seqindex = SeqIO.index('data/astral-scopdom-seqres-gd-sel-gs-bib-40-1.75.fa',
'fasta', key_function=lambda x: x.split()[0])
hie = pickle.load(Path('data/train/scop40_1fold_hie.pkl').open('rb'))
scop = Scop(dir_path='data/scop', version='1.75')
tmpdir = Path(f'.{algo}')
tmpdir.mkdir(exist_ok=True)
auc_result = {}
for sf in tqdm(hie):
px_list = hie[sf]
if len(px_list) < 1:
continue
sid = random.sample(px_list, 1)[0]
record = seqindex[sid]
f_fasta = tmpdir/f'{sid}.fasta'
f_xml = tmpdir/f'{sid}.xml'
SeqIO.write(record, f_fasta.as_posix(), 'fasta')
try:
if algo == 'psiblast':
NcbipsiblastCommandline(query=f_fasta.as_posix(),
db='astral-scopdom-seqres-gd-sel-gs-bib-40-1.75',
num_threads=int(os.cpu_count()),
num_iterations=3,
evalue=999999,
outfmt=5,
out=f_xml.as_posix())()
elif algo == 'deltablast':
NcbideltablastCommandline(query=f_fasta.as_posix(),
db='astral-scopdom-seqres-gd-sel-gs-bib-40-1.75',
num_threads=int(os.cpu_count()),
num_iterations=3,
evalue=999999,
outfmt=5,
out=f_xml.as_posix())()
else:
raise ValueError(f'Invalid algorithm ({algo})')
except ApplicationError as e:
logging.error(e)
f_xml.unlink()
continue
finally:
f_fasta.unlink()
results = SearchIO.parse(f_xml.as_posix(), 'blast-xml')
results = list(results)[-1]
results = list(results)[:500]
sf_sccs = scop.getNodeBySunid(sf).sccs
roc_score = []
roc_label = []
for result in results:
result_sf_sccs = result.description.split(' ')[0][:-2]
roc_score.append(-result.hsps[0].evalue)
if result_sf_sccs == sf_sccs:
roc_label.append(1)
else:
roc_label.append(0)
if np.all(np.array(roc_label) == 1):
auc = 1.0
elif np.all(np.array(roc_label) == 0):
auc = 0.0
else:
auc = metrics.roc_auc_score(roc_label, roc_score)
auc_result[sf_sccs] = {'auc': auc, 'sample': sid, 'num': len(results)}
f_xml.unlink()
now = int(time.time())
pickle.dump(auc_result, Path(f'auc_result_{algo}_{now}.pkl').open('wb'))
('d3cw9a1', datetime.datetime(2008, 9, 2, 0, 0), 'e.23.1', 22),
('d2axtd1', datetime.datetime(2009, 2, 10, 0, 0), 'f.26.1', 174),
('d2axto1', datetime.datetime(2009, 2, 10, 0, 0), 'f.4.1', 15),
('d2vy4a1', datetime.datetime(2009, 2, 17, 0, 0), 'g.37.1', 182),
('d3d9ta1', datetime.datetime(2009, 2, 10, 0, 0), 'g.52.1', 81),
)
test_data = [x[0] for x in test_data]
scop40 = SeqIO.index('data/astral-scopdom-seqres-gd-sel-gs-bib-40-1.75.fa', 'fasta')
scop100_hie = Scop(dir_path=Path('data/scop'), version='1.75')
if n_splits > 1:
fold1 = next(KFold(n_splits=n_splits, shuffle=True).split(scop40))
samples = np.array([v for i, v in enumerate(scop40) if i in fold1[1]])
# sf_sunid of scop100 (sid)
px = np.array([x.sid for x in scop100_hie.getNodeBySunid(sf_sunid).getDescendents('px')])
# select only sf_sunid in scop40 from scop100
isect = np.intersect1d(samples, px)
# select half of sf_sunid only in scop40 for test data
fold1 = next(KFold(n_splits=test_n_splits, shuffle=True).split(isect))
tests = np.array([v for i, v in enumerate(isect) if i in fold1[1]])
np.save(Path(f'data/test/scop40_{n_splits}fold_sf{sf_sunid}_testdata_{test_n_splits}fold.npy'), tests)
# select domain sids only in scop40 for training data
train = np.setdiff1d(samples, tests)
np.save(Path(f'data/train/scop40_{n_splits}fold_trainingdata.npy'), train)
# {sf: [sid]} list for making alignment pairs in the same superfamily
hie = {}
for i in train:
dom = scop100_hie.getDomainBySid(i)
if dom:
sf = dom.getAscendent('sf').sccs
