File size: 12,616 Bytes
1f88cea | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 | from dataclasses import dataclass
from pathlib import Path
from typing import Dict, List
import re
import numpy as np
import pytorch_lightning as pl
import torch
from torch import Tensor
from torch.utils.data import DataLoader
from rdkit.Chem import Mol
from boltzgen.data.data import (
Input,
Structure,
)
from boltzgen.data.feature.featurizer import Featurizer
from boltzgen.data.pad import pad_to_max
from boltzgen.data.mol import (
load_canonicals,
load_molecules,
)
from boltzgen.data.template.features import load_dummy_templates
from boltzgen.data.tokenize.tokenizer import Tokenizer
from boltzgen.data.template.features import (
load_dummy_templates,
)
@dataclass
class DataConfig:
"""Data configuration."""
target_dir: str
msa_dir: str
moldir: str
min_len: int
max_len: int
target_ids: str
tokenizer: Tokenizer
featurizer: Featurizer
backbone_only: bool = False
atom14: bool = False
atom37: bool = False
design: bool = False
target_structure_condition: bool = False
multiplicity: int = 1
disulfide_prob: float = 1.0
disulfide_on: bool = False
@dataclass
class Dataset:
"""Data holder."""
struct_dir: Path
record_dir: Path
target_ids: List[str]
min_len: int
max_len: int
tokenizer: Tokenizer
featurizer: Featurizer
multiplicity: int = 1
def collate(data: List[Dict[str, Tensor]]) -> Dict[str, Tensor]:
"""Collate the data.
Parameters
----------
data : List[Dict[str, Tensor]]
The data to collate.
Returns
-------
Dict[str, Tensor]
The collated data.
"""
# Get the keys
keys = data[0].keys()
# Collate the data
collated = {}
for key in keys:
values = [d[key] for d in data]
if key not in [
"all_coords",
"all_resolved_mask",
"crop_to_all_atom_map",
"chain_symmetries",
"amino_acids_symmetries",
"ligand_symmetries",
"activity_name",
"activity_qualifier",
"sid",
"cid",
"aid",
"normalized_protein_accession",
"pair_id",
"record",
"id",
"tokenized",
"structure",
"structure_bonds",
"extra_mols",
"data_sample_idx",
]:
# Check if all have the same shape
shape = values[0].shape
if not all(v.shape == shape for v in values):
values = pad_to_max(values, 0)
else:
values = torch.stack(values, dim=0)
# Stack the values
collated[key] = values
return collated
class PredictionDataset(torch.utils.data.Dataset):
"""Base iterable dataset."""
def __init__(
self,
dataset: Dataset,
canonicals: dict[str, Mol],
moldir: str,
backbone_only: bool = False,
atom14: bool = False,
atom37: bool = False,
design: bool = False,
target_structure_condition: bool = False,
disulfide_prob: float = 1.0,
disulfide_on: bool = False,
) -> None:
"""Initialize the training dataset.
Parameters
----------
datasets : List[Dataset]
The datasets to sample from.
"""
super().__init__()
self.dataset = dataset
self.moldir = moldir
self.canonicals = canonicals
self.backbone_only = backbone_only
self.atom14 = atom14
self.design = design
self.target_structure_condition = target_structure_condition
self.atom37 = atom37
self.disulfide_prob = disulfide_prob
self.disulfide_on = disulfide_on
def __getitem__(self, idx: int) -> Dict:
"""Get an item from the dataset.
Returns
-------
Dict[str, Tensor]
The sampled data features.
"""
# Get a sample from the dataset
data_sample_idx = idx // len(self.dataset.target_ids)
target_id = self.dataset.target_ids[idx % len(self.dataset.target_ids)]
# Get the structure
try:
target_id = target_id.lower()
ccd = target_id.split("_")[0]
pdb_id = target_id.split("_")[1].lower()
str_native = Structure.load(self.dataset.struct_dir / f"{pdb_id}.npz")
res_selection = str_native.residues["name"] == ccd.upper()
if res_selection.sum() > 1:
print(f"CCD {ccd} occurs multiple times in {pdb_id}. Using first.")
res_selection = np.arange(len(res_selection))[res_selection][[0]]
str_target = Structure.extract_residues(str_native, res_selection)
if self.dataset.min_len == self.dataset.max_len:
length = self.dataset.min_len
else:
length = np.random.random_integers(
self.dataset.min_len, self.dataset.max_len
)
str_prot = Structure.empty_protein(seq_len=length)
structure = Structure.concatenate(str_prot, str_target)
except Exception as e: # noqa: BLE001
print(f"Failed to load input for {target_id} with error {e}. Skipping.") # noqa: T201
return self.__getitem__(0)
# Check if the structure is valid
try:
tokenized = self.dataset.tokenizer.tokenize(structure)
except Exception as e: # noqa: BLE001
print(f"Tokenizer failed on {target_id} with error {e}. Skipping.") # noqa: T201
return self.__getitem__(0)
# Design mask logic
tokenized.tokens["design_mask"] = (
tokenized.tokens["asym_id"] == tokenized.tokens["asym_id"][0]
).astype(tokenized.tokens["asym_id"].dtype)
# Propagate design mask to obtain chain_design_mask (True whenever something is covalently bound to any residue that is in a chain that contains a design residue).
chain_design_mask = tokenized.tokens["design_mask"].astype(bool)
asym_id = tokenized.tokens["asym_id"]
while True:
design_chains = np.unique(asym_id[chain_design_mask])
chain_propagated = np.isin(asym_id, design_chains)
for i, j, _ in tokenized.bonds:
if any([chain_propagated[i], chain_propagated[j]]):
chain_propagated[i] = True
chain_propagated[j] = True
if np.equal(chain_propagated, chain_design_mask).all():
break
chain_design_mask = chain_propagated.astype(bool)
try:
# Try to find molecules in the dataset moldir if provided
# Find missing ones in global moldir and check if all found
molecules = {}
molecules.update(self.canonicals)
mol_names = set(tokenized.tokens["res_name"].tolist())
mol_names = mol_names - set(self.canonicals.keys())
if self.moldir is not None:
molecules.update(load_molecules(self.moldir, mol_names))
mol_names = mol_names - set(molecules.keys())
molecules.update(load_molecules(self.moldir, mol_names))
except Exception as e: # noqa: BLE001
print(f"Molecule loading failed for {target_id} with error {e}. Skipping.")
return self.__getitem__(0)
# Finalize input data
input_data = Input(
tokens=tokenized.tokens,
bonds=tokenized.bonds,
token_to_res=tokenized.token_to_res,
structure=structure,
msa={},
templates=None,
)
# Compute features
try:
features = self.dataset.featurizer.process(
input_data,
molecules=molecules,
random=np.random.default_rng(None),
training=False,
max_seqs=1,
backbone_only=self.backbone_only,
atom14=self.atom14,
atom37=self.atom37,
design=self.design,
override_method="X-RAY DIFFRACTION",
disulfide_prob=self.disulfide_prob,
disulfide_on=self.disulfide_on,
)
except Exception as e: # noqa: BLE001
print(f"Featurizer failed on {target_id} with error {e}. Skipping.") # noqa: T201
return self.__getitem__(0)
# set chain_design_mask
features["chain_design_mask"] = torch.from_numpy(chain_design_mask)
# Compute template features
templates_features = load_dummy_templates(
tdim=1, num_tokens=len(features["res_type"])
)
features.update(templates_features)
features["idx_dataset"] = torch.tensor(1)
def sanitize_filename(name):
return re.sub(r"[^A-Za-z0-9._-]", "_", name)
features["id"] = sanitize_filename(target_id)
if self.dataset.multiplicity > 1:
features["data_sample_idx"] = data_sample_idx
return features
def __len__(self) -> int:
"""Get the length of the dataset.
Returns
-------
int
The length of the dataset.
"""
return len(self.dataset.target_ids) * self.dataset.multiplicity
class LigandBinderDataModule(pl.LightningDataModule):
"""DataModule for BoltzGen."""
def __init__(self, cfg: DataConfig, batch_size, num_workers, pin_memory) -> None:
"""Initialize the DataModule.
Parameters
----------
config : DataConfig
The data configuration.
"""
super().__init__()
self.cfg = cfg
self.batch_size = batch_size
self.num_workers = num_workers
self.pin_memory = pin_memory
with Path(cfg.target_ids).open("r") as f:
target_ids = [x for x in f.read().splitlines()]
print("split", target_ids)
dataset = Dataset(
struct_dir=Path(cfg.target_dir) / "structures",
record_dir=Path(cfg.target_dir) / "records",
target_ids=target_ids,
multiplicity=cfg.multiplicity,
min_len=cfg.min_len,
max_len=cfg.max_len,
tokenizer=cfg.tokenizer,
featurizer=cfg.featurizer,
)
# Load canonical molecules
canonicals = load_canonicals(cfg.moldir)
self.predict_set = PredictionDataset(
dataset=dataset,
canonicals=canonicals,
moldir=Path(cfg.moldir),
backbone_only=cfg.backbone_only,
atom14=cfg.atom14,
design=cfg.design,
target_structure_condition=cfg.target_structure_condition,
disulfide_prob=cfg.disulfide_prob,
disulfide_on=cfg.disulfide_on,
)
def predict_dataloader(self) -> DataLoader:
"""Get the training dataloader.
Returns
-------
DataLoader
The training dataloader.
"""
return DataLoader(
self.predict_set,
batch_size=self.batch_size,
num_workers=self.num_workers,
pin_memory=self.pin_memory,
shuffle=False,
collate_fn=collate,
)
def transfer_batch_to_device(
self,
batch: Dict,
device: torch.device,
dataloader_idx: int, # noqa: ARG002
) -> Dict:
"""Transfer a batch to the given device.
Parameters
----------
batch : Dict
The batch to transfer.
device : torch.device
The device to transfer to.
dataloader_idx : int
The dataloader index.
Returns
-------
np.Any
The transferred batch.
"""
for key in batch:
if key not in [
"all_coords",
"all_resolved_mask",
"crop_to_all_atom_map",
"chain_symmetries",
"amino_acids_symmetries",
"ligand_symmetries",
"activity_name",
"activity_qualifier",
"sid",
"cid",
"normalized_protein_accession",
"pair_id",
"record",
"id",
"tokenized",
"structure",
"structure_bonds",
"extra_mols",
"data_sample_idx",
]:
batch[key] = batch[key].to(device)
return batch
|