File size: 3,743 Bytes
ca7299e |
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 |
# Copyright 2021 AlQuraishi Laboratory
#
# 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.
from Bio.SVDSuperimposer import SVDSuperimposer
import numpy as np
import torch
def _superimpose_np(reference, coords):
"""
Superimposes coordinates onto a reference by minimizing RMSD using SVD.
Args:
reference:
[N, 3] reference array
coords:
[N, 3] array
Returns:
A tuple of [N, 3] superimposed coords and the final RMSD.
"""
sup = SVDSuperimposer()
sup.set(reference, coords)
sup.run()
return sup
def _superimpose_single(reference, coords):
reference_np = reference.detach().cpu().numpy()
coords_np = coords.detach().cpu().numpy()
sup = _superimpose_np(reference_np, coords_np)
rot, tran = sup.get_rotran()
superimposed, rmsd = sup.get_transformed(), sup.get_rms()
return coords.new_tensor(superimposed), coords.new_tensor(rmsd), rot, tran
def superimpose(reference, coords, mask, return_transform=False):
"""
Superimposes coordinates onto a reference by minimizing RMSD using SVD.
Args:
reference:
[*, N, 3] reference tensor
coords:
[*, N, 3] tensor
mask:
[*, N] tensor
Returns:
A tuple of [*, N, 3] superimposed coords and [*] final RMSDs.
"""
def select_unmasked_coords(coords, mask):
return torch.masked_select(
coords,
(mask > 0.)[..., None],
).reshape(-1, 3)
batch_dims = reference.shape[:-2]
flat_reference = reference.reshape((-1,) + reference.shape[-2:])
flat_coords = coords.reshape((-1,) + reference.shape[-2:])
flat_mask = mask.reshape((-1,) + mask.shape[-1:])
superimposed_list = []
rmsds = []
rots = []
trans = []
for r, c, m in zip(flat_reference, flat_coords, flat_mask):
r_unmasked_coords = select_unmasked_coords(r, m)
c_unmasked_coords = select_unmasked_coords(c, m)
superimposed, rmsd, rot, tran = _superimpose_single(
r_unmasked_coords,
c_unmasked_coords
)
rots.append(rot)
trans.append(tran)
# This is very inelegant, but idk how else to invert the masking
# procedure.
count = 0
superimposed_full_size = torch.zeros_like(r)
for i, unmasked in enumerate(m):
if(unmasked):
superimposed_full_size[i] = superimposed[count]
count += 1
superimposed_list.append(superimposed_full_size)
rmsds.append(rmsd)
superimposed_stacked = torch.stack(superimposed_list, dim=0)
rmsds_stacked = torch.stack(rmsds, dim=0)
rots_stacked = torch.tensor(np.stack(rots, axis=0), device=coords.device)
trans_stacked = torch.tensor(np.stack(trans, axis=0), device=coords.device)
superimposed_reshaped = superimposed_stacked.reshape(
batch_dims + coords.shape[-2:]
)
rmsds_reshaped = rmsds_stacked.reshape(
batch_dims
)
if return_transform:
return superimposed_reshaped, rmsds_reshaped, rots_stacked, trans_stacked
return superimposed_reshaped, rmsds_reshaped
|