-
Notifications
You must be signed in to change notification settings - Fork 1
/
evalueate_valnet.py
185 lines (156 loc) · 6.79 KB
/
evalueate_valnet.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
import os
import sys
import argparse
import shutil
import numpy as np
import pandas as pd
import torch
import torch.utils.tensorboard
import seaborn as sns
sns.set_theme(style="darkgrid")
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
from scipy import stats
from torch.nn.utils import clip_grad_norm_
from torch_geometric.loader import DataLoader
from torch_geometric.transforms import Compose
from tqdm.auto import tqdm
import sys
sys.path.append(os.path.abspath('./'))
import utils.misc as misc
import utils.train as utils_train
import utils.transforms as trans
from datasets import get_dataset
from datasets.pl_data import FOLLOW_BATCH
from models.molopt_score_model import ScorePosNet3D
def get_auroc(y_true, y_pred, feat_mode):
y_true = np.array(y_true)
y_pred = np.array(y_pred)
avg_auroc = 0.
possible_classes = set(y_true)
for c in possible_classes:
auroc = roc_auc_score(y_true == c, y_pred[:, c])
avg_auroc += auroc * np.sum(y_true == c)
mapping = {
'basic': trans.MAP_INDEX_TO_ATOM_TYPE_ONLY,
'add_aromatic': trans.MAP_INDEX_TO_ATOM_TYPE_AROMATIC,
'full': trans.MAP_INDEX_TO_ATOM_TYPE_FULL
}
return avg_auroc / len(y_true)
def get_pearsonr(y_true, y_pred):
y_true = np.array(y_true)
y_pred = np.array(y_pred)
return stats.pearsonr(y_true, y_pred)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--config', type=str, default='./configs/training.yml')
parser.add_argument('--device', type=str, default='cuda')
parser.add_argument('--logdir', type=str, default='./logs_diffusion')
parser.add_argument('--ckpt', type=str, default='./logs_diffusion/training_2023_09_29__14_08_47/checkpoints/1415000.pt')
parser.add_argument('--tag', type=str, default='')
parser.add_argument('--value_only', action='store_true')
parser.add_argument('--train_report_iter', type=int, default=200)
args = parser.parse_args()
# load ckpt
if args.ckpt:
ckpt = torch.load(args.ckpt, map_location=args.device)
config = ckpt['config']
else:
# Load configs
config = misc.load_config(args.config)
config_name = os.path.basename(args.config)[:os.path.basename(args.config).rfind('.')]
misc.seed_all(config.train.seed)
# Transforms
protein_featurizer = trans.FeaturizeProteinAtom()
ligand_featurizer = trans.FeaturizeLigandAtom(config.data.transform.ligand_atom_mode)
transform_list = [
protein_featurizer,
ligand_featurizer,
trans.FeaturizeLigandBond(),
trans.NormalizeVina(config.data.name)
]
if config.data.transform.random_rot:
transform_list.append(trans.RandomRotation())
transform = Compose(transform_list)
# Datasets and loaders
dataset, subsets = get_dataset(
config=config.data,
transform=transform,
)
train_set, test_set = subsets['train'], subsets['test']
# follow_batch = ['protein_element', 'ligand_element']
collate_exclude_keys = ['ligand_nbh_list']
train_iterator = utils_train.inf_iterator(DataLoader(
train_set,
batch_size=config.train.batch_size,
shuffle=True,
num_workers=config.train.num_workers,
follow_batch=FOLLOW_BATCH,
exclude_keys=collate_exclude_keys
))
val_loader = DataLoader(test_set, config.train.batch_size, shuffle=False,
follow_batch=FOLLOW_BATCH, exclude_keys=collate_exclude_keys)
# Model
model = ScorePosNet3D(
config.model,
protein_atom_feature_dim=protein_featurizer.feature_dim,
ligand_atom_feature_dim=ligand_featurizer.feature_dim
).to(args.device)
# Optimizer and scheduler
optimizer = utils_train.get_optimizer(config.train.optimizer, model)
scheduler = utils_train.get_scheduler(config.train.scheduler, optimizer)
start_it = 0
if args.ckpt:
model.load_state_dict(ckpt['model'])
optimizer.load_state_dict(ckpt['optimizer'])
scheduler.load_state_dict(ckpt['scheduler'])
start_it = ckpt['iteration']
def validate(it):
# fix time steps
sum_loss, sum_loss_pos, sum_loss_v, sum_loss_exp, sum_n = 0, 0, 0, 0, 0
all_pred_v, all_true_v , all_pred_exp, all_true_exp= [], [], [], []
with torch.no_grad():
model.eval()
for batch in tqdm(val_loader, desc='Validate'):
batch = batch.to(args.device)
batch_size = batch.num_graphs
for t in np.linspace(0, model.num_timesteps - 1, 10).astype(int):
time_step = torch.tensor([t] * batch_size).to(args.device)
results = model.get_diffusion_loss(
protein_pos=batch.protein_pos,
protein_v=batch.protein_atom_feature.float(),
affinity=batch.affinity.float(),
batch_protein=batch.protein_element_batch,
ligand_pos=batch.ligand_pos,
ligand_v=batch.ligand_atom_feature_full,
batch_ligand=batch.ligand_element_batch,
time_step=time_step
)
loss, loss_pos, loss_v, loss_exp, pred_exp = results['loss'], results['loss_pos'], results['loss_v'], results['loss_exp'], results['pred_exp']
sum_loss += float(loss) * batch_size
sum_loss_pos += float(loss_pos) * batch_size
sum_loss_v += float(loss_v) * batch_size
sum_loss_exp += float(loss_exp) * batch_size
sum_n += batch_size
all_pred_v.append(results['ligand_v_recon'].detach().cpu().numpy())
all_true_v.append(batch.ligand_atom_feature_full.detach().cpu().numpy())
all_pred_exp.append(pred_exp.detach().cpu().numpy())
all_true_exp.append(batch.affinity.float().detach().cpu().numpy())
avg_loss = sum_loss / sum_n
avg_loss_pos = sum_loss_pos / sum_n
avg_loss_v = sum_loss_v / sum_n
avg_loss_exp = sum_loss_exp / sum_n
atom_auroc = get_auroc(np.concatenate(all_true_v), np.concatenate(all_pred_v, axis=0),
feat_mode=config.data.transform.ligand_atom_mode)
exp_pearsonr = get_pearsonr(np.concatenate(all_true_exp, axis=0), np.concatenate(all_pred_exp, axis=0))
data=pd.DataFrame({
'predict': np.concatenate(all_pred_exp, axis=0),
'ground truth': np.concatenate(all_true_exp, axis=0)
})
if args.value_only:
return avg_loss_exp
return avg_loss
try:
validate(0)
except KeyboardInterrupt:
print('Terminating...')