def load_process_DAVIS(path='./data',
binary=False,
convert_to_log=True,
threshold=30):
print('Beginning Processing...')
if not os.path.exists(path):
os.makedirs(path)
url = 'https://drive.google.com/uc?export=download&id=14h-0YyHN8lxuc0KV3whsaSaA-4KSmiVN'
saved_path = wget.download(url, path)
print('Beginning to extract zip file...')
with ZipFile(saved_path, 'r') as zip:
zip.extractall(path=path)
affinity = pd.read_csv(path + '/DAVIS/affinity.txt', header=None, sep=' ')
with open(path + '/DAVIS/target_seq.txt') as f:
target = json.load(f)
with open(path + '/DAVIS/SMILES.txt') as f:
drug = json.load(f)
target = list(target.values())
drug = list(drug.values())
SMILES = []
Target_seq = []
y = []
for i in range(len(drug)):
for j in range(len(target)):
SMILES.append(drug[i])
Target_seq.append(target[j])
y.append(affinity.values[i, j])
if binary:
print(
'Default binary threshold for the binding affinity scores are 30, you can adjust it by using the "threshold" parameter'
)
y = [1 if i else 0 for i in np.array(y) < threshold]
else:
if convert_to_log:
print('Default set to logspace (nM -> p) for easier regression')
y = convert_y_unit(np.array(y), 'nM', 'p')
else:
y = y
print('Done!')
return np.array(SMILES), np.array(Target_seq), np.array(y)
def process_BindingDB(path=None,
df=None,
y='Kd',
binary=False,
convert_to_log=True,
threshold=30):
if not os.path.exists(path):
os.makedirs(path)
if df is not None:
print('Loading Dataset from the pandas input...')
else:
print('Loading Dataset from path...')
df = pd.read_csv(path, sep='\t', error_bad_lines=False)
print('Beginning Processing...')
df = df[df[
'Number of Protein Chains in Target (>1 implies a multichain complex)']
== 1.0]
df = df[df['Ligand SMILES'].notnull()]
if y == 'Kd':
idx_str = 'Kd (nM)'
elif y == 'IC50':
idx_str = 'IC50 (nM)'
elif y == 'Ki':
idx_str = 'Ki (nM)'
elif y == 'EC50':
idx_str = 'EC50 (nM)'
else:
print('select Kd, Ki, IC50 or EC50')
df_want = df[df[idx_str].notnull()]
df_want = df_want[['BindingDB Reactant_set_id', 'Ligand InChI', 'Ligand SMILES',\
'PubChem CID', 'UniProt (SwissProt) Primary ID of Target Chain',\
'BindingDB Target Chain Sequence', idx_str]]
df_want.rename(columns={
'BindingDB Reactant_set_id': 'ID',
'Ligand SMILES': 'SMILES',
'Ligand InChI': 'InChI',
'PubChem CID': 'PubChem_ID',
'UniProt (SwissProt) Primary ID of Target Chain': 'UniProt_ID',
'BindingDB Target Chain Sequence': 'Target Sequence',
idx_str: 'Label'
},
inplace=True)
df_want['Label'] = df_want['Label'].str.replace('>', '')
df_want['Label'] = df_want['Label'].str.replace('<', '')
df_want['Label'] = df_want['Label'].astype(float)
# have at least uniprot or pubchem ID
df_want = df_want[df_want.PubChem_ID.notnull()
| df_want.UniProt_ID.notnull()]
df_want = df_want[df_want.InChI.notnull()]
df_want = df_want[df_want.Label <= 10000000.0]
print('There are ' + str(len(df_want)) + ' drug target pairs.')
if binary:
print(
'Default binary threshold for the binding affinity scores are 30, you can adjust it by using the "threshold" parameter'
)
y = [1 if i else 0 for i in df_want.Label.values < threshold]
else:
if convert_to_log:
print('Default set to logspace (nM -> p) for easier regression')
y = convert_y_unit(df_want.Label.values, 'nM', 'p')
else:
y = df_want.Label.values
return df_want.SMILES.values, df_want['Target Sequence'].values, np.array(
y)