/
predict.py
150 lines (118 loc) · 4.77 KB
/
predict.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
import argparse
import numpy as np
import os
from os.path import join, isfile
import torch
import torch.nn.functional as F
from dataset import get_input
from models.attentivedist import AttentiveDist
import util
def load_model(args, mode):
if mode == "attn":
attention = True
num_blocks = 5
blocks_after_attn = 40
else:
attention = False
num_blocks = 40
blocks_after_attn = 0
model = AttentiveDist(in_channels=151, out_channels_dist=20, out_channels_angle=72, channels=64, num_blocks=num_blocks, dropout=0.1,
dilation_list=[1,2,4], pool='Max', out_channels_omega=25 ,out_channels_theta=25 ,out_channels_phi=13,
attention=attention, blocks_after_attn=blocks_after_attn)
if args.cuda:
model.cuda("cuda")
device = torch.device('cuda')
else:
device = torch.device('cpu')
if mode == "attn":
checkpoint = torch.load(join(args.trained_models,"model_"+str(mode)),map_location=device)
model.load_state_dict(checkpoint['state_dict'])
else:
model.load_state_dict(torch.load(join(args.trained_models,"model_"+str(mode)),map_location=device))
model.eval()
return model
def pred(args, target, model, mode):
model.eval()
target_dir = join(args.input_dir, target)
input_features = get_input(target, target_dir)
if args.cuda:
input_features["10-3"] = input_features["10-3"].cuda()
input_features["10-1"] = input_features["10-1"].cuda()
input_features["1"] = input_features["1"].cuda()
input_features["10"] = input_features["10"].cuda()
if mode == "attn":
dist_outs,omega_outs,theta_outs,orientation_phi_outs,phi_outs,psi_outs,attention_maps = model(input_features["10-3"],input_features["10-1"],input_features["1"],input_features["10"])
attention_maps = attention_maps.data.cpu()
else:
dist_outs,omega_outs,theta_outs,orientation_phi_outs,phi_outs,psi_outs = model(input_features[mode])
#Distance
dist_outs = F.softmax(0.5 * (dist_outs + dist_outs.transpose(2,3)), dim=1)
dist_pred = np.array(dist_outs.data.cpu())
dist_pred = dist_pred[0]
#Backbone angles
phi_outs = F.softmax(phi_outs,dim=1)
phi_pred = np.array(phi_outs.data.cpu())
phi_pred = phi_pred[0]
psi_outs = F.softmax(psi_outs,dim=1)
psi_pred = np.array(psi_outs.data.cpu())
psi_pred = psi_pred[0]
#Orientation angles
omega_outs = F.softmax(0.5 * (omega_outs + omega_outs.transpose(2,3)), dim=1)
omega_pred = np.array(omega_outs.data.cpu())
omega_pred = omega_pred[0]
theta_outs = F.softmax(theta_outs,dim=1)
theta_pred = np.array(theta_outs.data.cpu())
theta_pred = theta_pred[0]
orientation_phi_outs = F.softmax(orientation_phi_outs,dim=1)
orientation_phi_pred = np.array(orientation_phi_outs.data.cpu())
orientation_phi_pred = orientation_phi_pred[0]
pred = {}
pred["dist"] = dist_pred
pred["phi"] = phi_pred
pred["psi"] = psi_pred
pred["omega"] = omega_pred
pred["theta"] = theta_pred
pred["orientation_phi"] = orientation_phi_pred
if mode == "attn":
pred["attention_maps"] = attention_maps
return pred
def pred_attentivedist(args, models, target):
pred_attn = pred(args, target, models["attn"], mode="attn")
pred_10_3 = pred(args, target, models["10-3"], mode="10-3")
pred_10_1 = pred(args, target, models["10-1"], mode="10-1")
pred_1 = pred(args, target, models["1"], mode="1")
pred_10 = pred(args, target, models["10"], mode="10")
pred_attentive_dist = {}
for k in pred_attn:
if k == 'attention_maps':
pred_attentive_dist['attention_maps'] = pred_attn['attention_maps']
else:
pred_attentive_dist[k] = (pred_attn[k] + pred_10_3[k] + pred_10_1[k] + pred_1[k] + pred_10[k])/5
np.savez(join(args.out, target + '_prediction'),
dist=pred_attentive_dist["dist"],
phi=pred_attentive_dist["phi"],
psi=pred_attentive_dist["psi"],
omega=pred_attentive_dist["omega"],
theta=pred_attentive_dist["theta"],
orientation_phi=pred_attentive_dist["orientation_phi"],
attention_maps=pred_attentive_dist["attention_maps"])
print("AttentiveDist prediction done for %s"%(target))
if __name__=="__main__":
parser = argparse.ArgumentParser(description='Use trained model to predict')
parser.add_argument('--target', type=str, required=True, default="", help='target protein name')
parser.add_argument('--trained_models', type=str, default="./trained_models", help='Model to load')
parser.add_argument('--input_dir', type=str, default="./input", help='directory containing features')
parser.add_argument('--out', type=str, default="./", help='directory to save the output')
parser.add_argument("--cuda", action='store_true', help="Use GPU for prediction")
args = parser.parse_args()
target = args.target
modes = ["attn", "10-3", "10-1", "1", "10"]
models = {}
for mode in modes:
models[mode] = load_model(args, mode)
if isfile(target):
targets = util.read_target_list(target)
else:
targets = [target]
for target in targets:
pred_attentivedist(args, models, target)