-
Notifications
You must be signed in to change notification settings - Fork 7
Expand file tree
/
Copy pathtask.py
More file actions
283 lines (240 loc) · 13.1 KB
/
task.py
File metadata and controls
283 lines (240 loc) · 13.1 KB
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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
from copy import deepcopy
from typing import Optional, Any
import numpy as np
import torch
from torch import nn
from torch.nn import functional as F
from torch_scatter import scatter_mean
from torchdrug import core, tasks, layers
from torchdrug.core import Registry as R
from tqdm import tqdm
from diffpack import rotamer, layer
from diffpack.rotamer import atom_name_vocab, bb_atom_name, res_sym_atom_posn, _rmsd_per_residue, _get_symm_atoms
from diffpack.schedule import SO2VESchedule
@R.register("tasks.TorsionalDiffusion")
class TorsionalDiffusion(tasks.Task, core.Configurable):
"""
NewTorsionalDiffusion is a class for simulating the torsional diffusion of a protein model.
It inherits from the tasks.Task and core.Configurable classes and uses these to setup and control the diffusion
simulation.
Attributes:
eps (float): A small number to avoid division by zero errors.
_option_members (set): A set containing the names of class attributes.
model (nn.Module): The neural network model to be used.
schedule_1pi_periodic (SO2VESchedule): The schedule for 1pi periodic tasks.
schedule_2pi_periodic (SO2VESchedule): The schedule for 2pi periodic tasks.
num_mlp_layer (int): The number of layers in the model.
graph_construction_model (Optional[Any]): The model used for graph construction.
verbose (int): Verbosity level.
train_chi_id (Optional[Any]): Chi angle for training ranging from 0 to 3. If not specified, random chi angles are trained.
"""
NUM_CHI_ANGLES = 4
eps = 1e-10
_option_members = {"task", "criterion", "metric"}
def __init__(self, sigma_embedding: nn.Module,
model: nn.Module,
torsion_mlp_hidden_dims: list,
schedule_1pi_periodic: SO2VESchedule,
schedule_2pi_periodic: SO2VESchedule,
graph_construction_model: Optional[Any] = None,
verbose: int = 0,
train_chi_id: Optional[Any] = None, ):
super(TorsionalDiffusion, self).__init__()
self.torsion_mlp_hidden_dims = torsion_mlp_hidden_dims
self.model_list = nn.ModuleList([deepcopy(model) for _ in range(self.NUM_CHI_ANGLES)])
self.sigma_embedding_list = nn.ModuleList([deepcopy(sigma_embedding) for _ in range(self.NUM_CHI_ANGLES)])
self.torsion_mlp_list = nn.ModuleList([layers.MLP(self.model_list[i].output_dim, torsion_mlp_hidden_dims
+ [4,]) for i in range(self.NUM_CHI_ANGLES)])
self.schedule_2pi_periodic = schedule_2pi_periodic
self.schedule_1pi_periodic = schedule_1pi_periodic
self.graph_construction_model = graph_construction_model
self.verbose = verbose
self.train_chi_id = train_chi_id
def forward(self, batch):
all_loss = torch.tensor(0, dtype=torch.float32, device=self.device)
metric = {}
# Sample from the schedule
protein = batch['graph']
t = self.schedule_1pi_periodic.sample_train_t(shape=(protein.batch_size,)).to(self.device)
# Add noise to protein
train_chi_id = np.random.randint(self.NUM_CHI_ANGLES) if self.train_chi_id is None else self.train_chi_id
batch = self.add_noise(batch, t, chi_id=train_chi_id)
# Predict and take loss
pred = self.predict(batch, all_loss, metric)
target = self.target(batch)
metric = self.evaluate(pred, target)
all_loss += metric["diffusion loss"]
return all_loss, metric
def add_noise(self, batch, t, chi_id=None):
"""Add noise to protein and update protein
Args:
batch (dict): batch
t (torch.Tensor): [num_graph] random number in [0, 1]
Returns:
batch (dict): dict with the following attributes:
Protein:
chi_1pi_periodic_mask (torch.Tensor): [num_residue, 4] bool
chi_2pi_periodic_mask (torch.Tensor): [num_residue, 4] bool
chi_mask (torch.Tensor): [num_residue, 4] bool
chi_id (int): chi angle to be trained
sigma (torch.Tensor): [num_graph] sigma
score (torch.Tensor): [num_residue, 4] score
"""
protein = batch['graph']
if chi_id is not None:
protein = rotamer.remove_by_chi(protein, chi_id)
chis = rotamer.get_chis(protein, protein.node_position) # [num_residue, 4]
# Add noise to chis
chis, score_1pi = self.schedule_1pi_periodic.add_noise(chis, t, protein.chi_1pi_periodic_mask)
chis, score_2pi = self.schedule_2pi_periodic.add_noise(chis, t, protein.chi_2pi_periodic_mask)
score = torch.where(protein.chi_1pi_periodic_mask, score_1pi, score_2pi)
protein = rotamer.set_chis(protein, chis) # TODO：maybe have bug
batch['protein'] = protein
batch['chi_id'] = chi_id
batch['sigma'] = self.schedule_1pi_periodic.t_to_sigma(t)
batch['score'] = score
return batch
def predict(self, batch, all_loss=None, metric=None):
protein = batch['graph']
chi_id = batch['chi_id']
sigma = batch['sigma'] # [num_graph]
if self.graph_construction_model:
protein = self.graph_construction_model(protein)
# Model forward
node_sigma = sigma[protein.atom2graph] # [num_node]
node_feature = self.sigma_embedding_list[chi_id](protein.node_feature.float(), node_sigma)
node_feature = self.model_list[chi_id](protein, node_feature, all_loss=all_loss, metric=metric)["node_feature"]
residue_feature = scatter_mean(node_feature, protein.atom2residue, dim=0, dim_size=protein.num_residue)
pred = self.torsion_mlp_list[chi_id](residue_feature)
# Scaled by norm
torsion_sigma = sigma[protein.residue2graph].unsqueeze(-1).expand(-1, self.NUM_CHI_ANGLES) # [num_residue, 4]
score_norm_1pi = torch.tensor(self.schedule_1pi_periodic.score_norm(torsion_sigma), device=self.device)
score_norm_2pi = torch.tensor(self.schedule_2pi_periodic.score_norm(torsion_sigma), device=self.device)
score_norm = torch.where(protein.chi_1pi_periodic_mask, score_norm_1pi, score_norm_2pi)
pred_score = pred * score_norm.sqrt()
# Mask out non-related chis
pred_score = pred_score * protein.chi_mask.to(pred_score.dtype)
return pred_score, score_norm
def target(self, batch):
protein = batch["graph"]
target_score = batch['score'] # Move to protein attribute
target_score = target_score * protein.chi_mask
return target_score
def evaluate(self, pred, target):
metric = {}
pred_score, score_norm = pred
target_score = target
metric["diffusion loss"] = ((target_score - pred_score) ** 2 / (score_norm + self.eps)).mean()
metric["diffusion base loss"] = (pred_score ** 2 / (score_norm + self.eps)).mean()
return metric
@torch.no_grad()
def generate(self, batch, randomize=True):
protein = batch['graph']
if randomize:
protein = rotamer.randomize(protein)
schedule = self.schedule_1pi_periodic.reverse_t_schedule.to(self.device)
for chi_id in tqdm(range(self.NUM_CHI_ANGLES), desc="Autoregressive generation"):
for j in range(len(schedule) - 1):
t = schedule[j]
dt = schedule[j] - schedule[j + 1] if j + 1 < len(schedule) else 1
chis = rotamer.get_chis(protein, protein.node_position) # [num_residue, 4]
# Predict score
sigma = self.schedule_1pi_periodic.t_to_sigma(t).repeat(protein.batch_size)
chi_protein = rotamer.remove_by_chi(protein, chi_id)
pred_score, _ = self.predict({
"graph": chi_protein,
"sigma": sigma,
"chi_id": chi_id
})
# Step backward
chis = self.schedule_1pi_periodic.step(chis, pred_score, t, dt, chi_protein.chi_1pi_periodic_mask)
chis = self.schedule_2pi_periodic.step(chis, pred_score, t, dt, chi_protein.chi_2pi_periodic_mask)
protein = rotamer.set_chis(protein, chis)
return batch
def get_metric(self, pred_protein, true_protein, metric):
# assert pred_pos.shape == true_pos.shape
pred_pos = pred_protein.node_position
true_pos = true_protein.node_position
protein = true_protein
pred_pos_per_residue = torch.zeros(protein.num_residue, len(atom_name_vocab), 3, device=protein.device)
true_pos_per_residue = torch.zeros(protein.num_residue, len(atom_name_vocab), 3, device=protein.device)
pred_pos_per_residue[protein.atom2residue, protein.atom_name] = pred_pos
true_pos_per_residue[protein.atom2residue, protein.atom_name] = true_pos
symm_true_pos_per_residue = _get_symm_atoms(true_pos_per_residue, protein.residue_type)
# Symmetric alignment
rmsd_per_residue = _rmsd_per_residue(pred_pos_per_residue, true_pos_per_residue, protein.sidechain37_mask)
sym_rmsd_per_residue = _rmsd_per_residue(pred_pos_per_residue, symm_true_pos_per_residue,
protein.sidechain37_mask)
sym_replace_mask = rmsd_per_residue > sym_rmsd_per_residue
rmsd_per_residue[sym_replace_mask] = sym_rmsd_per_residue[sym_replace_mask]
true_pos_per_residue[sym_replace_mask] = symm_true_pos_per_residue[sym_replace_mask]
true_pos = true_pos_per_residue[protein.atom2residue, protein.atom_name]
metric["atom_rmsd_per_residue"] = rmsd_per_residue
pred_chi = rotamer.get_chis(protein, pred_pos)
true_chi = rotamer.get_chis(protein, true_pos)
chi_diff = (pred_chi - true_chi).abs()
chi_ae = torch.minimum(chi_diff, 2 * np.pi - chi_diff)
chi_ae_periodic = torch.minimum(chi_ae, np.pi - chi_ae)
chi_ae[protein.chi_1pi_periodic_mask] = chi_ae_periodic[protein.chi_1pi_periodic_mask]
metric["chi_ae_rad"] = chi_ae[protein.chi_mask] # [num_residue, 4]
metric["chi_ae_deg"] = chi_ae[protein.chi_mask] * 180 / np.pi # [num_residue, 4]
for i in range(self.NUM_CHI_ANGLES):
metric[f"chi_{i}_ae_rad"] = chi_ae[:, i][protein.chi_mask[:, i]]
metric[f"chi_{i}_ae_deg"] = chi_ae[:, i][protein.chi_mask[:, i]] * 180 / np.pi
return metric
@R.register("tasks.ConfidencePrediction")
class ConfidencePrediction(TorsionalDiffusion):
eps = 1e-10
_option_members = {"task", "criterion", "metric"}
def __init__(self, sigma_embedding: nn.Module,
model: nn.Module,
confidence_model: nn.Module,
torsion_mlp_hidden_dims: list,
schedule_1pi_periodic: SO2VESchedule,
schedule_2pi_periodic: SO2VESchedule,
num_sample: int = 5,
num_mlp_layer: int = 1,
graph_construction_model: Optional[Any] = None,
verbose: int = 0,
train_chi_id: Optional[Any] = None):
super().__init__(sigma_embedding,
model,
torsion_mlp_hidden_dims,
schedule_1pi_periodic,
schedule_2pi_periodic,
graph_construction_model,
verbose,
train_chi_id)
self.confidence_model = confidence_model
self.num_sample = num_sample
self.mlp = layers.MLP(self.confidence_model.output_dim,
[self.confidence_model.output_dim] * num_mlp_layer + [1])
def predict_rmsd(self, batch, all_loss=None, metric=None):
protein = batch['graph']
if self.graph_construction_model:
protein = self.graph_construction_model(protein)
atom_feature = self.confidence_model(protein, protein.node_feature.float())["node_feature"]
residue_feature = scatter_mean(atom_feature, protein.atom2residue, dim=0,
dim_size=protein.num_residue) # [num_residue, feature_dim]
pred = self.mlp(residue_feature).squeeze(-1) # [num_residue]
return pred
@torch.no_grad()
def generate(self, batch, randomize=True):
protein = batch['graph']
if randomize:
protein = rotamer.randomize(protein)
best_protein = protein.clone()
best_rmsd = torch.zeros(protein.num_residue, device=self.device) + 1e6
for _ in tqdm(range(self.num_sample), desc="Confidence sampling"):
batch = super().generate(batch, randomize=True) # TODO: do we need to randomize?
protein = batch['graph']
rmsd = self.predict_rmsd(batch)
residue_update_mask = rmsd < best_rmsd # [num_residue]
atom_update_mask = residue_update_mask[protein.atom2residue] # [num_atom]
best_protein.node_position[atom_update_mask] = protein.node_position[atom_update_mask]
best_rmsd[residue_update_mask] = rmsd[residue_update_mask]
best_batch = {
"graph": best_protein,
"rmsd": best_rmsd
}
return best_batch