def _calculate_gene_start_end(df):
df['start'] = 0
df['end'] = 0
for index, row in df.iterrows():
if row.db == 'hg19':
exons = exon_interval_trees()[row.chromosome][row.cut_position]
for exon in exons:
if (exon.strand == '+') == (row.Strand == 'sense'):
gene = read_gencode()[
(read_gencode().gene_id == exon.gene_id) &
(read_gencode().feature == 'gene')]
if len(gene) != 1:
print('found more than one gene for ID: {}'.format(
gene))
gene = gene.iloc[0]
df[index].start = gene.start
df[index].end = gene.end
break
else:
print('no gene found for row {}'.format(row))
if row.db == 'mm9':
strand = '+' if row.Strand == 'sense' else '-'
start, end, gene_ids = mouse_gene_intervals()[row.chromosome + strand][row]
if len(gene_ids) != 1:
print('error: found more than one gene: {}, row: {}'.format(gene, row))
df[index].start = start
df[index].end = end
def score(gene_id, guides):
'''
Call the azimuth module model
:guides: A dataframe with all relevant information for the guides
TODO add amino acid cut position and percent peptides
:returns: a list of scores
'''
if len(guides) == 0:
return []
gene = read_gencode()[read_gencode().gene_id == gene_id].iloc[0]
cut_positions = guides.apply(lambda row:
('hg19', gene.seqname, row['cut_position']),
axis=1)
conservation_scores = process_dataframe(cut_positions)
# now build the azimuth-style feature data
# TODO Strand might fail again..
df = pd.DataFrame({
'Sequence':
guides.target,
'Target':
'dummy',
'Target gene':
'dummy',
'score_drug_gene_rank': [0.5] * len(guides), # no need
'drug':
'dummy',
'30mer':
guides.context,
'Strand': [
'sense' if orientation == 'FWD' else 'antisense'
for orientation in guides.orientation
],
'Percent Peptide':
guides.percent_peptide,
'Amino Acid Cut position':
guides.aa_cut_position
})
Xdf = df[['Sequence', 'Target', 'drug', '30mer',
'Strand']].set_index(['Sequence', 'Target', 'drug'])
gene_position = df[[
'Sequence', 'Target gene', 'drug', 'Percent Peptide',
'Amino Acid Cut position'
]].set_index(['Sequence', 'Target gene', 'drug'])
Y = df[['Sequence', 'Target gene', 'drug', 'score_drug_gene_rank'
]].set_index(['Sequence', 'Target gene', 'drug'])
conservation_scores.set_index(Xdf.index, inplace=True)
combined_features, y, genes, feature_names = extract_features(
Xdf, Y, gene_position, conservation_scores, order=1)
transformed_features = feature_scaler().transform(combined_features)
result = cnn38_model()(Variable(
torch.from_numpy(transformed_features))).cpu().data.numpy()
return result.reshape(-1)
def main():
mismatches = {}
overflow_count = 0
gene_ids = read_gencode().gene_id.drop_duplicates()
if COMPUTATION_CORES > 1:
with Pool(COMPUTATION_CORES) as pool:
for partial_overflow_count, partial_mismatches in tqdm(
pool.imap_unordered(generate_exon_guides, gene_ids),
total=len(gene_ids)):
overflow_count += partial_overflow_count
for key in partial_mismatches:
try:
mismatches[key] += partial_mismatches[key]
except KeyError:
mismatches[key] = partial_mismatches[key]
else:
# debuggable
for gene_id in tqdm(gene_ids, total=len(gene_ids)):
partial_overflow_count, partial_mismatches = generate_exon_guides(
gene_id)
overflow_count += partial_overflow_count
for key in partial_mismatches:
try:
mismatches[key] += partial_mismatches[key]
except KeyError:
mismatches[key] = partial_mismatches[key]
# save anaysis data
print('Overflow count: {}'.format(overflow_count))
print(mismatches)
def gene_names_similar(gene_a, gene_b):
gc = read_gencode()
name_a = gc[gc.gene_id == gene_a].iloc[0].gene_name
name_b = gc[gc.gene_id == gene_b].iloc[0].gene_name
if name_a[:-1] in name_b or name_b[:-1] in name_a:
return True
else:
return False
def pdbs_for_gene(gene_id):
gencode = read_gencode()
pdbs = pdb_list()
protein_ids = gencode.loc[(gencode['gene_id'] == gene_id
)]['swissprot_id'].drop_duplicates().dropna()
canonical_pids = np.unique(
[normalize_pid(pid) for pid in protein_ids if pid]).astype('O')
if len(canonical_pids) > 1:
logging.warning('Gene {} has {} "canonical" protein ids: {}"'.format(
gene_id, len(canonical_pids), canonical_pids))
gene_pdbs = pdbs.loc[pdbs.SP_PRIMARY.isin(canonical_pids)][[
'PDB', 'SP_PRIMARY', 'CHAIN'
]].copy()
gene_pdbs.columns = ['pdb', 'swissprot_id', 'chain']
if len(gene_pdbs) > 0:
# mappings from swissprot-coordinate to pdb-index
gene_pdbs['mappings'] = gene_pdbs.apply(
lambda row: pdb_mappings(row.pdb, row.chain, row.swissprot_id),
axis=1)
empty_mappings = gene_pdbs['mappings'].apply(len) == 0
if empty_mappings.any():
logging.warning('No PDB mapping for {}'.format(
gene_pdbs[empty_mappings].pdb))
gene_pdbs.drop(gene_pdbs.index[empty_mappings], inplace=True)
gene_pdbs['start'] = gene_pdbs['mappings'].apply(
lambda mappings: min(mappings.values()))
gene_pdbs['end'] = gene_pdbs['mappings'].apply(
lambda mappings: max(mappings.values()) + 1) # end is contained
gene_pdbs['mappings'] = gene_pdbs['mappings'].apply(
lambda mappings:
{str(key): value
for key, value in mappings.items()})
return gene_pdbs
def test_read_gencode(mocked_gtf, mocked_protein_mapping):
'''only
Mocking read_gtf_as_dataframe to return a simplified dataframe,
read_gencode should return only one copy of gene2
'''
df = data.read_gencode()
# check that the correct columns are returned
eq_(
set(df.columns), {
'feature', 'gene_id', 'transcript_id', 'start', 'end', 'exon_id',
'exon_number', 'gene_name', 'transcript_type', 'strand',
'gene_type', 'tag', 'protein_id', 'swissprot_id', 'score',
'seqname', 'source'
})
# gene 'GB' should have zero entires for ENSEMBL
# but 3 (gene, transcript, exon) for HAVANA
eq_(len(df[(df.gene_id == 'GB') & (df.source == 'ENSEMBL')]), 0)
eq_(len(df[(df.gene_id == 'GB') & (df.source == 'HAVANA')]), 3)
# gene 'GC' should not be existent because it has no basic transcripts
eq_(len(df[df.gene_id == 'GC']), 0)
# start should be subtracted (just testing for one exon for simplicity)
eq_(df[df.exon_id == 'EB1'].iloc[0].start, 4)
# end should remain the same
eq_(df[df.exon_id == 'EB1'].iloc[0].end, 10)
# swissprot should be merged
eq_(list(df[(df.protein_id == 'PA')].swissprot_id.drop_duplicates()),
['SP1'])
eq_(list(df[(df.protein_id == 'PB')].swissprot_id.drop_duplicates()),
['SP2'])
# no swissprot_id for a protein_id shouldn't delete that row
assert (df.protein_id == 'PA2').any()