-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_trvate.py
226 lines (178 loc) · 9.7 KB
/
data_trvate.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
import torch
import pickle
import os
import pandas as pd
import numpy as np
import dgllife.utils as chemutils
from torch.utils.data import Dataset
from rdkit.Chem import AllChem
from tqdm import tqdm
data_dir_prefix = './'
hp = {
# data restriction (not change)
"pos_prec": ['[M+H]+', '[M+H-H2O]+', '[M+H-2H2O]+', '[M+H-NH3]+', '[M+Na]+', '[M+H+2i]+'],
"neg_prec": ['[M-H]-', '[M-H-H2O]-', '[M-H-CO2]-'],
"element_list": "chnopsh",
"data_dir": 'final_5',
"mode": 'positive',
"atom_feature": 'medium',
"bond_feature": 'light',
"ms_transformation": 'log10over3',
"max_mz": 1000,
"instrument_on_node": True,
"self_loop": True,
"num_virtual_nodes": 0,
"fp_size": 4096,
"noise": False,
# bin size (change)
"resolution": 1}
class msgnnDataset(Dataset):
def __init__(self, data_list, noise):
self.data_list = data_list
self.noise = noise
def __len__(self):
return len(self.data_list)
def __getitem__(self, idx):
i, pmz, g, setting_tensor, ms, fp = self.data_list[idx]
return g, setting_tensor, ms, pmz, fp
def get_atom_featurizer(feature_mode, element_list):
atom_mass_fun = chemutils.ConcatFeaturizer([chemutils.atom_mass])
def atom_type_one_hot(atom):
return chemutils.atom_type_one_hot(atom, allowable_set=element_list, encode_unknown=True)
if feature_mode == 'medium':
atom_featurizer_funs = chemutils.ConcatFeaturizer([
chemutils.atom_mass,
atom_type_one_hot,
chemutils.atom_total_degree_one_hot,
chemutils.atom_total_num_H_one_hot,
chemutils.atom_is_aromatic_one_hot,
chemutils.atom_is_in_ring_one_hot])
return chemutils.BaseAtomFeaturizer({"h": atom_featurizer_funs, "m": atom_mass_fun})
def get_bond_featurizer(feature_mode, self_loop):
if feature_mode == 'light':
return chemutils.BaseBondFeaturizer(featurizer_funcs={'e': chemutils.ConcatFeaturizer([chemutils.bond_type_one_hot])}, self_loop=self_loop)
def get_ms_setting_all_nodes(precursor_type, ce, n_nodes, prec_pool):
out = torch.zeros((n_nodes, len(prec_pool) + 1))
out[:, prec_pool.index(precursor_type)] = 1.0
out[:, -1] = ce
return out
def get_ms_setting(precursor_type, ce, prec_pool):
out = np.zeros(len(prec_pool) + 1)
out[prec_pool.index(precursor_type)] = 1.0
out[-1] = ce
return out
def get_intensity_(x):
return list(map(float, x.split(' ', 2)[0:2]))
def get_intensity(x):
x_list = list(map(get_intensity_, x.split('\n')[:-1]))
return np.array(x_list, dtype = float)
def get_ms_array(x, transformation, max_mz, resolution):
mz_intensity = get_intensity(x)
n_cells = int(max_mz / resolution)
ms_array = np.zeros(n_cells, np.float32)
mz_intensity = [p for p in mz_intensity if p[0] < max_mz + 1]
for p in mz_intensity:
bin_idx = int((p[0] - 1) / resolution)
ms_array[bin_idx] += p[1]
if transformation == "log10over3":
out = np.log10(ms_array + 1) / 3
return out
def mod_instrg_df2list(df, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes, prec_types):
if element_list == "chnopsh":
ele_list = ['H', 'C', 'O', 'N', 'P', 'S', 'F', 'Cl', 'Br', 'I']
instru_g_tensor = {}
for index, rw in df.iterrows():
g = chemutils.mol_to_bigraph(
mol_dict[rw['InChIKey']],
node_featurizer = get_atom_featurizer(atom_feature, ele_list),
edge_featurizer=get_bond_featurizer(edge_feature, self_loop),
add_self_loop = self_loop,
num_virtual_nodes = num_virtual_nodes
)
setting_tensor_on_nodes = get_ms_setting_all_nodes(rw['Precursor_type'], rw['NCE'], g.num_nodes(), prec_types)
instru_g_tensor[rw['InChIKey']] = setting_tensor_on_nodes
return setting_tensor_on_nodes
def df2list(df, trans, max_mz, resolution, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes, prec_types, instrument_on_node, fp_size):
if element_list == "chnopsh":
ele_list = ['H', 'C', 'O', 'N', 'P', 'S', 'F', 'Cl', 'Br', 'I']
data_list = []
for index, rw in df.iterrows():
setting_tensor = get_ms_setting(rw['Precursor_type'], rw['NCE'], prec_types)
fp = np.array([int(x) for x in AllChem.GetMorganFingerprintAsBitVect(mol_dict[rw['InChIKey']], radius = 2, nBits = fp_size).ToBitString()])
g = chemutils.mol_to_bigraph(
mol_dict[rw['InChIKey']],
node_featurizer = get_atom_featurizer(atom_feature, ele_list),
edge_featurizer=get_bond_featurizer(edge_feature, self_loop),
add_self_loop = self_loop,
num_virtual_nodes = num_virtual_nodes
)
if instrument_on_node:
setting_tensor_on_nodes = get_ms_setting_all_nodes(rw['Precursor_type'], rw['NCE'], g.num_nodes(), prec_types)
g.ndata['h'] = torch.cat((g.ndata['h'], setting_tensor_on_nodes), -1)
data_list.append((rw['InChIKey'], int(rw['PrecursorMZ']/resolution), g, setting_tensor, get_ms_array(rw['ms'], trans, max_mz, resolution), fp))
return data_list
def split_train_val(df, trans, max_mz, resolution, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes, prec_types, instrument_on_node, fp_size, ratio=0.9):
all_ik = pd.unique(df.InChIKey)
ik_msk = np.random.random(all_ik.shape)
train_ik = [ik for ik, msk in zip(all_ik, ik_msk) if msk < ratio]
val_ik = [ik for ik, msk in zip(all_ik, ik_msk) if msk >= ratio]
train_df = df[df.InChIKey.isin(train_ik)].reset_index(drop=True)
val_df = df[df.InChIKey.isin(val_ik)].reset_index(drop=True)
train_list = df2list(train_df, trans, max_mz, resolution, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes, prec_types, instrument_on_node, fp_size)
val_list = df2list(val_df, trans, max_mz, resolution, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes, prec_types,instrument_on_node, fp_size)
return train_list, val_list
def create_train_val_dataset(mode, data, precs, atom_feature, edge_feature, ms_transformation, max_mz, resolution,instrument_on_node, self_loop, num_virtual_nodes, element_list, fp_size, noise):
mol_dict, pos_train, neg_train = data
pos_prec, neg_prec = precs
if mode == 'positive':
prec_types = pos_prec
train_set, val_set = split_train_val(pos_train, ms_transformation, max_mz, resolution, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes,prec_types, instrument_on_node, fp_size, ratio=0.9)
train_ds = msgnnDataset(train_set, noise)
val_ds = msgnnDataset(val_set, noise)
return train_ds, val_ds
def create_test_dataset(mode, data, precs, atom_feature, edge_feature, ms_transformation, max_mz, resolution, instrument_on_node, self_loop, num_virtual_nodes, element_list, fp_size, noise):
mol_dict, pos_test, neg_test = data
pos_prec, neg_prec = precs
if mode == 'positive':
prec_types = pos_prec
test_set = df2list(pos_test, ms_transformation, max_mz, resolution, mol_dict, atom_feature, element_list, edge_feature, self_loop, num_virtual_nodes, prec_types, instrument_on_node, fp_size)
test_ds = msgnnDataset(test_set, noise)
return test_ds
def convert_graph_dgl_to_torchgeometric(ds):
for i in tqdm(range(len(ds))):
ds[i] = list(ds[i])
graph = ds[i][2]
g_nodes = graph.nodes().numpy()
g_edges = [x.numpy() for x in graph.edges()]
g_edges = np.vstack(g_edges)
g_edges_f = graph.edata['e'].numpy()
g_nodes_f = graph.ndata['h'].numpy()[g_nodes]
ds[i][2] = [g_edges, g_nodes_f, g_edges_f]
with open("./data/torch_trvate_"+str(int(1000/hp['resolution']))+"bin.pkl", 'wb') as fp:
pickle.dump(ds, fp, protocol=4)
def load_trvate(hp):
print('Loading Data...')
pos_train = pd.read_csv(os.path.join(data_dir_prefix+'data', hp['data_dir'], "pos_train.csv"))
neg_train = pd.read_csv(os.path.join(data_dir_prefix+'data', hp['data_dir'], "neg_train.csv"))
with open(os.path.join(data_dir_prefix+'data', hp['data_dir'], "mol_dict.pkl"), 'rb') as f:
mol_dict = pickle.load(f)
print('Creating Dataset...')
train_ds, val_ds = create_train_val_dataset(hp['mode'], (mol_dict, pos_train, neg_train), (hp['pos_prec'], hp['neg_prec']),
hp['atom_feature'], hp['bond_feature'], hp['ms_transformation'], hp['max_mz'], hp['resolution'],
hp['instrument_on_node'], hp['self_loop'], hp['num_virtual_nodes'], hp['element_list'], hp['fp_size'], hp['noise'])
print('Loading Data...')
pos_test = pd.read_csv(os.path.join(data_dir_prefix+'data', hp['data_dir'], "pos_test.csv"))
neg_test = pd.read_csv(os.path.join(data_dir_prefix+'data', hp['data_dir'], "neg_test.csv"))
print('Creating Dataset...')
test_ds = create_test_dataset(hp['mode'], (mol_dict, pos_test, neg_test), (hp['pos_prec'], hp['neg_prec']),
hp['atom_feature'], hp['bond_feature'], hp['ms_transformation'], hp['max_mz'], hp['resolution'],
hp['instrument_on_node'], hp['self_loop'], hp['num_virtual_nodes'], hp['element_list'], hp['fp_size'], hp['noise'])
tr_l, va_l, te_l = len(train_ds.data_list), len(val_ds.data_list), len(test_ds.data_list)
print(tr_l, va_l, te_l)
tr_idx, va_idx, te_idx = [i for i in range(tr_l)], [i for i in range(tr_l, tr_l+ va_l)], [i for i in range(tr_l+ va_l, tr_l+ va_l+ te_l)]
with open("./data/trvate_idx.pkl", 'wb') as f:
pickle.dump((tr_idx, va_idx, te_idx), f)
union_ds = list(train_ds.data_list) + list(val_ds.data_list) + list(test_ds.data_list)
convert_graph_dgl_to_torchgeometric(union_ds)
if __name__ == "__main__":
load_trvate(hp)