Spaces:
Sleeping
Sleeping
File size: 9,775 Bytes
ce7bf5b |
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 |
# Copyright Generate Biomedicines, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Layers for euclidean symmetry group operations
This module contains pytorch layers for symmetry operations for point groups (Cyclic, Dihedral, Tetrahedral, Octahedral and Icosahedral)
"""
import itertools
import math
from itertools import product
from typing import List, Optional, Tuple, Union
import numpy as np
import torch
from scipy.optimize import linear_sum_assignment
from scipy.spatial.transform import Rotation
from tqdm import tqdm
from chroma.layers.structure import backbone
TAU = 0.5 * (1 + math.sqrt(5))
ROT_DICT = {
"O": [
[[0.0, 1.0, 0.0], [-1.0, 0.0, 0.0], [0.0, 0.0, 1.0]],
[[0.0, 0.0, 1.0], [-1.0, 0.0, 0.0], [0.0, -1.0, 0.0]],
[[1.0, 0.0, 0.0], [0.0, 0.0, -1.0], [0.0, 1.0, 0.0]],
[[0.0, 0.0, 1.0], [0.0, -1.0, 0.0], [1.0, 0.0, 0.0]],
[[-1.0, 0.0, 0.0], [0.0, 0.0, 1.0], [0.0, 1.0, 0.0]],
[[0.0, -1.0, 0.0], [1.0, 0.0, 0.0], [0.0, 0.0, 1.0]],
[[0.0, 1.0, 0.0], [0.0, 0.0, 1.0], [1.0, 0.0, 0.0]],
[[0.0, 0.0, 1.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]],
[[-1.0, 0.0, 0.0], [0.0, -1.0, 0.0], [0.0, 0.0, 1.0]],
[[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]],
[[1.0, 0.0, 0.0], [0.0, 0.0, 1.0], [0.0, -1.0, 0.0]],
[[0.0, 0.0, 1.0], [0.0, 1.0, 0.0], [-1.0, 0.0, 0.0]],
[[0.0, -1.0, 0.0], [0.0, 0.0, 1.0], [-1.0, 0.0, 0.0]],
[[0.0, 1.0, 0.0], [0.0, 0.0, -1.0], [-1.0, 0.0, 0.0]],
[[0.0, 0.0, -1.0], [0.0, 1.0, 0.0], [1.0, 0.0, 0.0]],
[[0.0, 0.0, -1.0], [1.0, 0.0, 0.0], [0.0, -1.0, 0.0]],
[[0.0, 0.0, -1.0], [-1.0, 0.0, 0.0], [0.0, 1.0, 0.0]],
[[0.0, -1.0, 0.0], [0.0, 0.0, -1.0], [1.0, 0.0, 0.0]],
[[0.0, 1.0, 0.0], [1.0, 0.0, 0.0], [0.0, 0.0, -1.0]],
[[-1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, -1.0]],
[[1.0, 0.0, 0.0], [0.0, -1.0, 0.0], [0.0, 0.0, -1.0]],
[[0.0, -1.0, 0.0], [-1.0, 0.0, 0.0], [0.0, 0.0, -1.0]],
[[0.0, 0.0, -1.0], [0.0, -1.0, 0.0], [-1.0, 0.0, 0.0]],
[[-1.0, 0.0, 0.0], [0.0, 0.0, -1.0], [0.0, -1.0, 0.0]],
],
"T": [
[[-1.0, 0.0, 0.0], [0.0, -1.0, 0.0], [0.0, 0.0, 1.0]],
[[1.0, 0.0, 0.0], [0.0, -1.0, 0.0], [0.0, 0.0, -1.0]],
[[0.0, 0.0, -1.0], [1.0, 0.0, 0.0], [0.0, -1.0, 0.0]],
[[0.0, 0.0, -1.0], [-1.0, 0.0, 0.0], [0.0, 1.0, 0.0]],
[[0.0, 0.0, 1.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]],
[[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]],
[[-1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, -1.0]],
[[0.0, 1.0, 0.0], [0.0, 0.0, 1.0], [1.0, 0.0, 0.0]],
[[0.0, 0.0, 1.0], [-1.0, 0.0, 0.0], [0.0, -1.0, 0.0]],
[[0.0, -1.0, 0.0], [0.0, 0.0, 1.0], [-1.0, 0.0, 0.0]],
[[0.0, 1.0, 0.0], [0.0, 0.0, -1.0], [-1.0, 0.0, 0.0]],
[[0.0, -1.0, 0.0], [0.0, 0.0, -1.0], [1.0, 0.0, 0.0]],
],
}
def get_point_group(group: str) -> torch.Tensor:
"""get representation of group elements at torch.Tensor
Args:
group (str): group names, selecting from {"C_{n}" , "D_{n}", "T", "O", "I" }
Returns:
torch.Tensor: rotation matrices for queried point groups
"""
if group.startswith("C"):
n = group.split("_")[1]
G = get_Cn_groups(int(n))
elif group.startswith("D"):
n = group.split("_")[1]
G = get_Dn_groups(int(n))
elif group == "I":
G = get_I_rotations()
elif group == "O" or group == "T":
G = torch.Tensor(np.array(ROT_DICT[group]))
else:
raise ValueError("{ } not available".format(group))
return G
def get_Cn_groups(n: int) -> torch.Tensor:
"""get rotation matrices for Cyclic groups
Args:
n (int): symmetry order
Returns:
torch.Tensor: n x 3 x 3
"""
G = []
for ri in range(n):
cos_phi = np.round(np.cos(ri * np.pi * 2 / n), 4)
sin_phi = np.round(np.sin(ri * np.pi * 2 / n), 4)
g = np.array(
[[cos_phi, -sin_phi, 0.0], [sin_phi, cos_phi, 0.0], [0.0, 0.0, 1.0]]
)
G.append(np.round(g, 4))
return torch.Tensor(np.array(G))
def get_Dn_groups(n: int) -> torch.Tensor:
"""get rotation matrices for Dihedral groups
Args:
n (int): symmetry order
Returns:
torch.Tensor: 2n x 3 x 3
"""
cos_phi = np.round(
np.cos(np.pi * 2 / n), 8
) # unify the choice of # of decimals to keep
sin_phi = np.round(np.sin(np.pi * 2 / n), 8)
rot_generator = np.array(
[[cos_phi, -sin_phi, 0.0], [sin_phi, cos_phi, 0.0], [0.0, 0.0, 1.0]]
)
b = np.array([[1, 0, 0], [0, -1, 0], [0, 0, -1]])
G = []
c = rot_generator
for itr in range(n):
c_new = c @ rot_generator
G.append(c_new)
c = c_new
G.append(b @ c)
return torch.Tensor(np.array(G))
def get_I_rotations(tree_depth: int = 5) -> torch.Tensor:
"""get rotation matrices for the Icosahedral group (I)
Returns:
torch.Tensor: 60 x 3 x 3
"""
tree_depth = 5 # tree traverse depth
g1 = torch.Tensor([[-1.0, 0.0, 0.0], [0.0, -1.0, 0.0], [0.0, 0.0, 1.0]])
g2 = torch.Tensor([[0.0, 0.0, 1.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]])
g3 = torch.Tensor(
[
[0.5, -0.5 * TAU, 0.5 / TAU],
[0.5 * TAU, 0.5 / TAU, -0.5],
[0.5 / TAU, 0.5, 0.5 * TAU],
]
)
# g4 = torch.Tensor([[-1., 0., 0.], [0., -1., 0.], [0., 0., -1.]]) reflection operation
generators = [g1, g2, g3]
gen1 = generators
gen2 = generators
sym_ops = []
# todo: there is probably better way to do this
# brute force search by traversing the Caley graph
for itr in range(tree_depth):
mat_prod = [mat_pair[0] @ mat_pair[1] for mat_pair in product(gen1, gen2)]
sym_ops += mat_prod
gen1 = mat_prod
# find unique rotation matrices
sym_ops = torch.unique(torch.round(torch.stack(sym_ops, dim=0), decimals=6), dim=0)
return sym_ops
def subsample(
X: torch.Tensor,
C: torch.Tensor,
G: torch.Tensor,
knbr: int,
seed_idx: Optional[int] = None,
):
"""generate substructures based on distances between subunit COM
Args:
X (torch.Tensor): structures
C (torch.Tensor): chain map
G (torch.Tensor): rotation matrices
knbr (int): number of nearest neighbors
seed_idx (int, optional): seed idx, this will be randomly selected if set to None. Defaults to None.
Returns:
tuple: substructure coordinates, chain map, indices associated with all the substructure chains, seed idx
"""
if knbr > G.shape[0] - 1:
knbr = G.shape[0] - 1
G_order = G.shape[0]
X_chain_com = X.reshape(1, G.shape[0], -1, 3).mean(-2)
if seed_idx is None:
seed_idx = torch.randint(0, G.shape[0], (1,)).item()
Dis_chain = (
(X_chain_com.unsqueeze(-2) - X_chain_com.unsqueeze(-3)).pow(2).sum(-1).sqrt()
)
subdomain_idx = Dis_chain[0, seed_idx].topk(knbr + 1, largest=False)[1]
X_subdomain = X.reshape(1, G.shape[0], -1, 4, 3)[:, subdomain_idx]
X_subdomain = X_subdomain.reshape(1, -1, 4, 3)
C_subdomain = C.reshape(1, G.shape[0], -1)[:, : knbr + 1, :].reshape(1, -1)
return X_subdomain, C_subdomain, subdomain_idx, seed_idx
def symmetrize_XCS(
X: torch.Tensor,
C: torch.LongTensor,
S: torch.LongTensor,
G: torch.Tensor,
device: str = "cpu",
) -> Tuple[torch.Tensor, torch.LongTensor, torch.LongTensor]:
"""Symmetrize a protein structure with a given symmetry group.
This function takes a data tensor X, a chain tensor C, a Sequence Tensor S, and a symmetry group tensor G. The function returns a symmetrized data tensor X_complex, a modified chain tensor C_complex, and a replicated Sequence Tensor S_complex. The function uses the torch.einsum function to apply the symmetry group G to each chain in X and concatenate them into X_complex. The function also modifies the chain labels in C by multiplying them by the symmetry index and concatenates them into C_complex. The function also replicates the Sequence Tensor S by the number of symmetry elements and concatenates them into S_complex.
Args:
X (torch.Tensor): Data tensor with shape `(batch_size, num_residues, 4, 3)`.
C (torch.LongTensor): Chain tensor with shape `(batch_size, num_residues)`.
S (torch.LongTensor): Sequence tensor with shape `(batch_size, num_residues)`.
G (torch.Tensor): Symmetry group tensor with shape `(n_sym, 3, 3)`.
device (str, optional): The device to use for computation. Defaults to "cpu".
Returns:
torch.Tensor: Symmetrized data tensor with shape `(batch_size, num_residues * n_sym, 4, 3)`.
torch.LongTensor: Modified chain tensor with shape `(batch_size, num_residues * n_sym)`.
torch.LongTensor: Modified Sequence tensor with shape `(batch_size * n_sym,)`.
"""
G = G.to(S.device)
X_complex = torch.einsum("gij,bnlj->gnli", G, X).reshape(1, -1, 4, 3).to(device)
C_complex = torch.cat([C * (i + 1) for i in range(G.shape[0])], 1).to(device)
S_complex = torch.cat([S for i in range(G.shape[0])], 1).to(device)
return X_complex, C_complex, S_complex
|