File size: 8,912 Bytes
db32e07 | 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 | from __future__ import annotations
import ctypes
import sys
from dataclasses import dataclass
from pathlib import Path
from typing import Iterable, List
import numpy as np
from numpy.lib.format import open_memmap
from tqdm import tqdm
from .embeddings import l2_normalize
@dataclass
class FLAREMolEmbedder:
flare_repo: str
hparams_pth: str
checkpoint_pth: str
device: str = "cpu"
batch_size: int = 256
normalize: bool = True
def _load(self):
libstdcpp = Path(sys.executable).resolve().parent.parent / "lib" / "libstdc++.so.6"
if libstdcpp.exists():
ctypes.CDLL(str(libstdcpp), mode=ctypes.RTLD_GLOBAL)
import dgl
import torch
import yaml
flare_repo = Path(self.flare_repo).resolve()
if str(flare_repo) not in sys.path:
sys.path.insert(0, str(flare_repo))
from flare.data.transforms import MolToGraph
from flare.utils.models import get_model
with open(self.hparams_pth) as f:
params = yaml.load(f, Loader=yaml.FullLoader)
params["checkpoint_pth"] = str(self.checkpoint_pth)
params["df_test_path"] = ""
params["accelerator"] = "cpu"
params["devices"] = 1
device = self.device
if device == "cuda":
if not torch.cuda.is_available():
device = "cpu"
else:
try:
dgl.graph(([0], [0])).to("cuda")
except Exception:
device = "cpu"
model = get_model(params["model"], params)
model = model.to(device)
model.eval()
mol_transform = MolToGraph(
atom_feature=params["atom_feature"],
bond_feature=params["bond_feature"],
element_list=params["element_list"],
)
return model.mol_enc_model, mol_transform, device
def encode(self, smiles: Iterable[str]) -> np.ndarray:
import dgl
import torch
smiles_list = list(smiles)
if not smiles_list:
return np.zeros((0, 0), dtype=np.float32)
mol_encoder, mol_transform, device = self._load()
outputs: List[np.ndarray] = []
total_batches = (len(smiles_list) + self.batch_size - 1) // self.batch_size
with torch.no_grad():
for start in tqdm(range(0, len(smiles_list), self.batch_size), total=total_batches, desc="Encoding FLARE", unit="batch"):
batch_smiles = smiles_list[start:start + self.batch_size]
graphs = [mol_transform(smi) for smi in batch_smiles]
batched = dgl.batch(graphs)
if device != "cpu":
batched = batched.to(device)
node_embeddings = mol_encoder(batched)
pooled = mol_encoder.pool(batched, node_embeddings)
outputs.append(pooled.detach().cpu().numpy().astype(np.float32))
arr = np.concatenate(outputs, axis=0).astype(np.float32)
if self.normalize:
arr = l2_normalize(arr)
return arr
def encode_to_npy(self, smiles: Iterable[str], out_path: str | Path) -> Path:
import dgl
import torch
smiles_list = list(smiles)
out_path = Path(out_path)
if not smiles_list:
np.save(out_path, np.zeros((0, 0), dtype=np.float32))
return out_path
mol_encoder, mol_transform, device = self._load()
total_batches = (len(smiles_list) + self.batch_size - 1) // self.batch_size
mm = None
offset = 0
with torch.no_grad():
for start in tqdm(range(0, len(smiles_list), self.batch_size), total=total_batches, desc="Encoding FLARE", unit="batch"):
batch_smiles = smiles_list[start:start + self.batch_size]
graphs = [mol_transform(smi) for smi in batch_smiles]
batched = dgl.batch(graphs)
if device != "cpu":
batched = batched.to(device)
node_embeddings = mol_encoder(batched)
pooled = mol_encoder.pool(batched, node_embeddings)
chunk = pooled.detach().cpu().numpy().astype(np.float32)
if mm is None:
mm = open_memmap(out_path, mode="w+", dtype=np.float32, shape=(len(smiles_list), chunk.shape[1]))
mm[offset:offset + chunk.shape[0]] = chunk
offset += chunk.shape[0]
if mm is None:
np.save(out_path, np.zeros((0, 0), dtype=np.float32))
return out_path
del mm
if self.normalize:
arr = np.load(out_path, mmap_mode="r+")
norms = np.linalg.norm(arr, axis=1, keepdims=True)
arr[:] = arr[:] / np.clip(norms, 1e-8, None)
del arr
return out_path
@dataclass
class FLARESpecEmbedder:
flare_repo: str
hparams_pth: str
checkpoint_pth: str
dataset_pth: str
subformula_dir_pth: str
fold: str = "test"
device: str = "cpu"
batch_size: int = 128
normalize: bool = True
def _load(self):
libstdcpp = Path(sys.executable).resolve().parent.parent / "lib" / "libstdc++.so.6"
if libstdcpp.exists():
ctypes.CDLL(str(libstdcpp), mode=ctypes.RTLD_GLOBAL)
import dgl
import torch
import yaml
flare_repo = Path(self.flare_repo).resolve()
if str(flare_repo) not in sys.path:
sys.path.insert(0, str(flare_repo))
from massspecgym.models.base import Stage
from flare.data.datasets import MassSpecDataset_PeakFormulas
from flare.utils.data import get_spec_featurizer
from flare.utils.models import get_model
with open(self.hparams_pth) as f:
params = yaml.load(f, Loader=yaml.FullLoader)
params["checkpoint_pth"] = str(self.checkpoint_pth)
params["df_test_path"] = ""
params["accelerator"] = "cpu"
params["devices"] = 1
device = self.device
if device == "cuda":
if not torch.cuda.is_available():
device = "cpu"
else:
try:
dgl.graph(([0], [0])).to("cuda")
except Exception:
device = "cpu"
model = get_model(params["model"], params)
model = model.to(device)
model.eval()
spec_transform = get_spec_featurizer(params["spectra_view"], params)
dataset = MassSpecDataset_PeakFormulas(
spectra_view=params["spectra_view"],
spec_transform=spec_transform,
mol_transform=None,
pth=self.dataset_pth,
subformula_dir_pth=self.subformula_dir_pth,
formula_source=params.get("formula_source", "default"),
return_mol_freq=False,
return_identifier=True,
stage=Stage.TEST,
)
return model.spec_enc_model, params["spectra_view"], dataset, device
def encode(self):
import torch
spec_encoder, spectra_view, dataset, device = self._load()
metadata = dataset.metadata
if "fold" in metadata.columns and self.fold:
metadata = metadata[metadata["fold"].astype(str) == str(self.fold)]
indices = metadata.index.to_list()
if not indices:
return np.zeros((0, 0), dtype=np.float32), [], []
outputs: List[np.ndarray] = []
out_smiles: List[str] = []
out_ids: List[str] = []
with torch.no_grad():
for start in tqdm(range(0, len(indices), self.batch_size), desc="Encoding FLARE spectra", unit="batch"):
batch_idx = indices[start:start + self.batch_size]
specs = []
n_peaks = []
for idx in batch_idx:
item = dataset.__getitem__(idx, transform_mol=False)
spec = item[spectra_view]
specs.append(spec)
n_peaks.append(int(spec.shape[0]))
row = dataset.metadata.loc[idx]
out_smiles.append(str(row["smiles"]))
out_ids.append(str(row["identifier"]))
batch = torch.nn.utils.rnn.pad_sequence(specs, batch_first=True, padding_value=-5)
batch = batch.to(device)
enc = spec_encoder(batch, n_peaks)
if enc.ndim == 3:
mask = (batch != -5).any(dim=-1).float()
pooled = (enc * mask.unsqueeze(-1)).sum(dim=1) / mask.sum(dim=1, keepdim=True).clamp(min=1.0)
else:
pooled = enc
outputs.append(pooled.detach().cpu().numpy().astype(np.float32))
arr = np.concatenate(outputs, axis=0).astype(np.float32)
if self.normalize:
arr = l2_normalize(arr)
return arr, out_smiles, out_ids
|