dualmapref 0.2.8__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (186) hide show
  1. args_parse.py +593 -0
  2. candidate_corrector.py +161 -0
  3. candidate_scorer.py +359 -0
  4. conformend/__init__.py +36 -0
  5. conformend/refiner.py +25 -0
  6. dataset/__init__.py +1 -0
  7. dataset/drugdataset.py +644 -0
  8. dataset/manifold.py +237 -0
  9. dataset/sharded_drugdataset.py +124 -0
  10. dataset/smiles2graph.py +217 -0
  11. dualmapref/__init__.py +24 -0
  12. dualmapref/refiner.py +16 -0
  13. dualmapref-0.2.8.dist-info/LICENSE +21 -0
  14. dualmapref-0.2.8.dist-info/METADATA +167 -0
  15. dualmapref-0.2.8.dist-info/RECORD +186 -0
  16. dualmapref-0.2.8.dist-info/WHEEL +5 -0
  17. dualmapref-0.2.8.dist-info/entry_points.txt +10 -0
  18. dualmapref-0.2.8.dist-info/top_level.txt +29 -0
  19. exputils.py +289 -0
  20. infer.py +23 -0
  21. infer_base_conformers.py +5589 -0
  22. infer_mean_dispersion_torsion.py +1117 -0
  23. infer_torsion.py +752 -0
  24. infer_torsional_denoiser.py +1021 -0
  25. manimol/__init__.py +30 -0
  26. manimol/__main__.py +6 -0
  27. manimol/cli.py +36 -0
  28. manimol/full_cli.py +41 -0
  29. manimol/legacy_imports.py +43 -0
  30. manimol/manifest.py +56 -0
  31. manimol/postprocess.py +148 -0
  32. manimol/priors.py +60 -0
  33. manimol/protocols.py +119 -0
  34. manimol/refiner.py +645 -0
  35. manimol/runtime.py +194 -0
  36. manimol/selection.py +85 -0
  37. manimol/torsion.py +22 -0
  38. manimol/train_cli.py +14 -0
  39. mixture_flow/src/__init__.py +1 -0
  40. mixture_flow/src/_path.py +7 -0
  41. mixture_flow/src/infer28_mixture_torsion_flow.py +1093 -0
  42. mixture_flow/src/mixture_torsion_flow.py +291 -0
  43. mixture_flow/src/summarize_mixture_debug.py +80 -0
  44. mixture_flow/src/train28_mixture_torsion_flow.py +583 -0
  45. mixture_torsion_flow.py +7 -0
  46. models/ manifold_learning.py +86 -0
  47. models/__init__.py +8 -0
  48. models/dist.py +273 -0
  49. models/dist2coords.py +23 -0
  50. models/egnn.py +107 -0
  51. models/experimental_backbones.py +364 -0
  52. models/gnnconv.py +512 -0
  53. models/kernel_inversion.py +141 -0
  54. models/kernels.py +427 -0
  55. models/losses.py +154 -0
  56. models/mean_dispersion_bridge.py +119 -0
  57. models/mean_dispersion_torsion_context.py +155 -0
  58. models/model.py +73 -0
  59. models/model0.py +2375 -0
  60. models/vis.py +128 -0
  61. pgraph_controls.py +107 -0
  62. scripts/__init__.py +2 -0
  63. scripts/analyze_basin_alignment.py +63 -0
  64. scripts/analyze_hif2a_conformer_aggregation.py +267 -0
  65. scripts/analyze_hif2a_contact_pharmacophore.py +1111 -0
  66. scripts/analyze_hif2a_fixed_budget.py +202 -0
  67. scripts/analyze_hif2a_pharmacophore.py +405 -0
  68. scripts/analyze_hif2a_pharmacophore_match_metric.py +386 -0
  69. scripts/analyze_p_control_candidate_flow.py +154 -0
  70. scripts/analyze_scaffold_basin.py +83 -0
  71. scripts/audit_dispersion_proposals.py +208 -0
  72. scripts/audit_etflow_protocol.py +258 -0
  73. scripts/audit_external_overlap.py +150 -0
  74. scripts/audit_oldbest_fulltrainval_split.py +82 -0
  75. scripts/audit_pdbbind_contact_pharmacophores.py +146 -0
  76. scripts/audit_qm9_geomol_test_split.py +94 -0
  77. scripts/audit_stage1_stage2_cross.py +131 -0
  78. scripts/build_geom_drugs_lowref_diagnostic.py +84 -0
  79. scripts/build_qm9_backfill_manifest.py +99 -0
  80. scripts/check_etflow_runtime.py +56 -0
  81. scripts/compare_prior_sources.py +122 -0
  82. scripts/compose_binding_mode_figure.py +87 -0
  83. scripts/compose_hif2a_contact_pharmacophore_suppfig.py +255 -0
  84. scripts/compose_hif2a_functional_geometry_pipeline.py +266 -0
  85. scripts/compose_manimol_internal_architecture.py +291 -0
  86. scripts/diagnose_flow_basin_intervention.py +54 -0
  87. scripts/diagnose_hif2a_pharmacophore_budget.py +185 -0
  88. scripts/diagnose_route2_p_ranking.py +123 -0
  89. scripts/discover_binding_mode_inputs.py +30 -0
  90. scripts/evaluate_functional_prior_stage1.py +230 -0
  91. scripts/evaluate_geomanimol.py +74 -0
  92. scripts/evaluate_geomanimol_flow.py +23 -0
  93. scripts/evaluate_geomanimol_multiscaffold.py +46 -0
  94. scripts/evaluate_geomanimol_residual.py +45 -0
  95. scripts/evaluate_geomanimol_torsion.py +40 -0
  96. scripts/evaluate_hif2a_bioactive_enrichment.py +91 -0
  97. scripts/evaluate_hif2a_pms_suite.py +440 -0
  98. scripts/evaluate_pdbbind_contact_pms.py +155 -0
  99. scripts/evaluate_platinum_external.py +114 -0
  100. scripts/evaluate_platinum_pool.py +82 -0
  101. scripts/evaluate_route2_calibrated_p.py +104 -0
  102. scripts/evaluate_route2_library_components.py +596 -0
  103. scripts/evaluate_route2_p_basin_budget.py +288 -0
  104. scripts/evaluate_stage1_checkpoint_only.py +30 -0
  105. scripts/export_dispersion_bond_audit.py +136 -0
  106. scripts/export_top20_shards.py +207 -0
  107. scripts/external_validation_metrics.py +152 -0
  108. scripts/fit_topology_prior.py +117 -0
  109. scripts/frozen_dispersion_audit.py +281 -0
  110. scripts/geomanimol_debug.py +50 -0
  111. scripts/geomanimol_smoke.py +37 -0
  112. scripts/infer_etflow_matched.py +524 -0
  113. scripts/inspect_checkpoint.py +12 -0
  114. scripts/inspect_conformer_capacity.py +71 -0
  115. scripts/inspect_pool.py +12 -0
  116. scripts/make_figure3_selection_loss.py +474 -0
  117. scripts/make_figure3_selection_summary.py +275 -0
  118. scripts/make_geom_debug_subset.py +62 -0
  119. scripts/make_low_reference_test_subset.py +70 -0
  120. scripts/merge_stage1_into_stage2.py +82 -0
  121. scripts/place_hif2a_candidates_for_evaluation.py +66 -0
  122. scripts/plot_hif2a_low_budget_efficiency.py +100 -0
  123. scripts/plot_prior_validation.py +48 -0
  124. scripts/precompute_stage2_assignments.py +90 -0
  125. scripts/prepare_bace_activity.py +98 -0
  126. scripts/prepare_bace_etkdg_benchmark.py +90 -0
  127. scripts/prepare_figure5_4x4_cases.py +470 -0
  128. scripts/prepare_geom_drugs_geomol_top20.py +472 -0
  129. scripts/prepare_pdbbind_core_raw.py +90 -0
  130. scripts/prepare_platinum_raw.py +74 -0
  131. scripts/prepare_pubchem_etkdg.py +54 -0
  132. scripts/rebuild_hif2a_final_sdf_and_metrics.py +55 -0
  133. scripts/rebuild_hif2a_fixed_pool_sweep.py +175 -0
  134. scripts/rebuild_hif2a_strict.py +220 -0
  135. scripts/recompute_hif2a_bootstrap.py +63 -0
  136. scripts/reference_free_geometry.py +112 -0
  137. scripts/render_binding_modes.py +413 -0
  138. scripts/render_figure5_4x4.py +330 -0
  139. scripts/render_figure6_pharmacophore_pymol.py +162 -0
  140. scripts/render_hif2a_contact_pharmacophore_pymol.py +414 -0
  141. scripts/render_hif2a_figure11.py +397 -0
  142. scripts/run_bace_2d_baseline.py +73 -0
  143. scripts/run_drugs_validation_matrix.py +225 -0
  144. scripts/run_external_reference_free.py +53 -0
  145. scripts/run_manimol_finalk_sensitivity.py +106 -0
  146. scripts/run_manimol_oversampling_sensitivity.py +109 -0
  147. scripts/run_pdisp_matched_intervention.py +116 -0
  148. scripts/run_stage1_stage2_cross20.py +83 -0
  149. scripts/score_hif2a_candidates_with_pgraph.py +221 -0
  150. scripts/summarize_dispersion_intervention.py +88 -0
  151. scripts/test_mean_dispersion_bridge.py +50 -0
  152. scripts/train_dispersion_head_diagnostic.py +320 -0
  153. scripts/train_dispersion_head_streaming.py +519 -0
  154. scripts/train_geomanimol.py +100 -0
  155. scripts/train_geomanimol_flow.py +26 -0
  156. scripts/train_geomanimol_residual.py +70 -0
  157. scripts/train_geomanimol_torsion.py +50 -0
  158. scripts/train_quantile_representation_diagnostic.py +204 -0
  159. scripts/validate_mean_dispersion_prior.py +173 -0
  160. scripts/validate_packed_shard.py +41 -0
  161. scripts/validate_stage1_prior.py +221 -0
  162. scripts/validate_top20_shards.py +36 -0
  163. scripts/validation/__init__.py +1 -0
  164. scripts/validation/hif2a_downstream_baselines.py +105 -0
  165. scripts/validation/hif2a_pairwise_ranker.py +218 -0
  166. scripts/verify_binding_mode_figure.py +47 -0
  167. torsion_manifold.py +281 -0
  168. train.py +78 -0
  169. train_graph_encoder.py +206 -0
  170. train_local_torsion_corrector.py +511 -0
  171. train_mean_dispersion_torsion_prior.py +329 -0
  172. train_probability_basin.py +235 -0
  173. train_torsion_generator.py +1631 -0
  174. train_torsion_manifold_prior.py +258 -0
  175. train_torsional_denoiser.py +415 -0
  176. trainer.py +1038 -0
  177. utils/__init__.py +7 -0
  178. utils/checkpoint.py +40 -0
  179. utils/device.py +17 -0
  180. utils/helpers.py +98 -0
  181. utils/kabsch.py +29 -0
  182. utils/lookup_table.py +164 -0
  183. utils/metrics.py +50 -0
  184. utils/optuna.py +111 -0
  185. utils/save_mol.py +366 -0
  186. utils/util.py +36 -0
dataset/drugdataset.py ADDED
@@ -0,0 +1,644 @@
1
+ import os
2
+ import pickle
3
+ from collections import OrderedDict
4
+
5
+ import torch
6
+ from torch_geometric.data import InMemoryDataset, Data
7
+ from rdkit import Chem, RDLogger
8
+ from rdkit.Geometry import Point3D
9
+ from tqdm import tqdm
10
+
11
+ from .smiles2graph import smile2graph4GEOM
12
+
13
+ # The debug/top20 files were serialized with a newer RDKit. Older RDKit
14
+ # emits a repetitive depickling compatibility notice although the records
15
+ # remain readable for this pipeline; suppress that non-actionable noise.
16
+ RDLogger.DisableLog("rdApp.*")
17
+
18
+
19
+ def _as_float_tensor(x):
20
+ if x is None:
21
+ return None
22
+ if torch.is_tensor(x):
23
+ return x.float()
24
+ return torch.tensor(x, dtype=torch.float)
25
+
26
+
27
+ def build_torsion_index(mol):
28
+ """
29
+ torsion_index shape:
30
+ [4, T]
31
+ """
32
+ torsions = []
33
+
34
+ if mol is None:
35
+ return torch.empty((4, 0), dtype=torch.long)
36
+
37
+ for bond in mol.GetBonds():
38
+ if bond.IsInRing():
39
+ continue
40
+
41
+ j = bond.GetBeginAtomIdx()
42
+ k = bond.GetEndAtomIdx()
43
+
44
+ nbr_j = [
45
+ atom.GetIdx()
46
+ for atom in mol.GetAtomWithIdx(j).GetNeighbors()
47
+ if atom.GetIdx() != k
48
+ ]
49
+
50
+ nbr_k = [
51
+ atom.GetIdx()
52
+ for atom in mol.GetAtomWithIdx(k).GetNeighbors()
53
+ if atom.GetIdx() != j
54
+ ]
55
+
56
+ for i in nbr_j:
57
+ for l in nbr_k:
58
+ if len({i, j, k, l}) == 4:
59
+ torsions.append([i, j, k, l])
60
+
61
+ if len(torsions) == 0:
62
+ return torch.empty((4, 0), dtype=torch.long)
63
+
64
+ return torch.tensor(
65
+ torsions,
66
+ dtype=torch.long,
67
+ ).t().contiguous()
68
+
69
+
70
+ def enrich_edge_attr_with_bond_flags(mol, edge_index, edge_attr):
71
+ """
72
+ append:
73
+ aromatic
74
+ conjugated
75
+ ring
76
+ """
77
+ if edge_index is None:
78
+ return edge_attr
79
+
80
+ extras = []
81
+
82
+ for eid in range(edge_index.size(1)):
83
+ u = int(edge_index[0, eid])
84
+ v = int(edge_index[1, eid])
85
+
86
+ bond = mol.GetBondBetweenAtoms(u, v)
87
+
88
+ if bond is None:
89
+ extras.append([0.0, 0.0, 0.0])
90
+ else:
91
+ extras.append([
92
+ float(bond.GetIsAromatic()),
93
+ float(bond.GetIsConjugated()),
94
+ float(bond.IsInRing()),
95
+ ])
96
+
97
+ extras = torch.tensor(extras, dtype=torch.float)
98
+
99
+ if edge_attr is None:
100
+ return extras
101
+
102
+ edge_attr = _as_float_tensor(edge_attr)
103
+
104
+ if edge_attr.dim() == 1:
105
+ edge_attr = edge_attr.view(-1, 1)
106
+
107
+ if edge_attr.size(0) != extras.size(0):
108
+ return extras
109
+
110
+ return torch.cat([edge_attr, extras], dim=-1)
111
+
112
+
113
+ class QM9Dataset(InMemoryDataset):
114
+ def __init__(
115
+ self,
116
+ name="QM9",
117
+ root="data",
118
+ transform=None,
119
+ pre_transform=None,
120
+ pre_filter=None,
121
+ max_confs=5,
122
+ raw_prefix="converted",
123
+ keep_full_test_conformers=False,
124
+ test_raw_filename=None,
125
+ train_raw_filename=None,
126
+ val_raw_filename=None,
127
+ process_train_mol_limit=0,
128
+ process_val_mol_limit=0,
129
+ process_test_mol_limit=0,
130
+ store_rdkit_objects=True,
131
+ ):
132
+ self.raw_prefix = str(raw_prefix)
133
+ self.name = name
134
+ self.root = root
135
+ self.requested_max_confs = int(max_confs)
136
+ self.keep_full_test_conformers = bool(keep_full_test_conformers)
137
+ self.test_raw_filename = str(test_raw_filename) if test_raw_filename else None
138
+ self.train_raw_filename = str(train_raw_filename) if train_raw_filename else None
139
+ self.val_raw_filename = str(val_raw_filename) if val_raw_filename else None
140
+ self.process_train_mol_limit = int(process_train_mol_limit)
141
+ self.process_val_mol_limit = int(process_val_mol_limit)
142
+ self.process_test_mol_limit = int(process_test_mol_limit)
143
+ self.store_rdkit_objects = bool(store_rdkit_objects)
144
+ # ``0`` is an explicit auto mode. We resolve it before PyG checks the
145
+ # processed-file path so caches from different conformer caps can never
146
+ # be silently reused.
147
+ self.max_confs = self._resolve_max_confs(self.requested_max_confs)
148
+
149
+ super().__init__(
150
+ root,
151
+ transform,
152
+ pre_transform,
153
+ pre_filter,
154
+ )
155
+
156
+ self.data, self.slices = torch.load(
157
+ self.processed_paths[0],
158
+ weights_only=False,
159
+ )
160
+
161
+ self.train_index, self.valid_index, self.test_index = pickle.load(
162
+ open(self.processed_paths[1], "rb")
163
+ )
164
+
165
+ @property
166
+ def raw_dir(self):
167
+ return self.root
168
+
169
+ @property
170
+ def raw_file_names(self):
171
+ if any((self.train_raw_filename, self.val_raw_filename, self.test_raw_filename)):
172
+ return [
173
+ self.train_raw_filename or f"{self.raw_prefix}_train.pt",
174
+ self.val_raw_filename or f"{self.raw_prefix}_val.pt",
175
+ self.test_raw_filename or f"{self.raw_prefix}_test.pt",
176
+ ]
177
+ if self.raw_prefix == "geom_qm9":
178
+ return [
179
+ "geom_qm9_train.pt",
180
+ "geom_qm9_val.pt",
181
+ "geom_qm9_test.pt",
182
+ ]
183
+ if self.raw_prefix == "geom_qm9_full":
184
+ return [
185
+ "geom_qm9_train_full.pt",
186
+ "geom_qm9_val_full.pt",
187
+ "geom_qm9_test_full.pt",
188
+ ]
189
+ if self.raw_prefix == "geom_drugs":
190
+ return [
191
+ "geom_drugs_train.pt",
192
+ "geom_drugs_val.pt",
193
+ "geom_drugs_test.pt",
194
+ ]
195
+ if self.raw_prefix == "geom_drugs_revisited":
196
+ return [
197
+ "geom_drugs_revisited_train.pt",
198
+ "geom_drugs_revisited_val.pt",
199
+ "geom_drugs_revisited_test.pt",
200
+ ]
201
+ if self.raw_prefix.startswith("geom_drugs75k"):
202
+ return [
203
+ f"{self.raw_prefix}_train.pt",
204
+ f"{self.raw_prefix}_val.pt",
205
+ f"{self.raw_prefix}_test.pt",
206
+ ]
207
+ return [
208
+ "train_converted_data1K.pt",
209
+ "val_converted_data100.pt",
210
+ "test_converted_data100.pt",
211
+ ]
212
+
213
+ @property
214
+ def processed_dir(self):
215
+ return os.path.join(self.root, "processed_data")
216
+
217
+ @property
218
+ def processed_file_names(self):
219
+ # Legacy defaults stay byte-for-byte compatible with the established
220
+ # best-method cache. Full-test retention is opt-in for new studies.
221
+ if self.max_confs == 5 and not self.keep_full_test_conformers:
222
+ suffix = ""
223
+ elif self.keep_full_test_conformers:
224
+ suffix = f"_v2_c{self.max_confs}_fulltest"
225
+ else:
226
+ suffix = f"_v2_c{self.max_confs}"
227
+ if self.process_train_mol_limit or self.process_val_mol_limit or self.process_test_mol_limit:
228
+ suffix += (f"_train{self.process_train_mol_limit}"
229
+ f"_val{self.process_val_mol_limit}"
230
+ f"_test{self.process_test_mol_limit}")
231
+ if self.train_raw_filename or self.val_raw_filename or self.test_raw_filename:
232
+ suffix += "_customraw"
233
+ if not self.store_rdkit_objects:
234
+ suffix += "_no_rdkit"
235
+ if self.raw_prefix == "geom_qm9":
236
+ return [
237
+ f"data_geom_qm9_probmanifold_conformer_unit{suffix}.pt",
238
+ f"split_geom_qm9_probmanifold_conformer_unit{suffix}.pt",
239
+ ]
240
+ if self.raw_prefix == "geom_qm9_full":
241
+ return [
242
+ f"data_geom_qm9_full_probmanifold_conformer_unit{suffix}.pt",
243
+ f"split_geom_qm9_full_probmanifold_conformer_unit{suffix}.pt",
244
+ ]
245
+ if self.raw_prefix == "geom_drugs":
246
+ return [
247
+ f"data_geom_drugs_probmanifold_conformer_unit{suffix}.pt",
248
+ f"split_geom_drugs_probmanifold_conformer_unit{suffix}.pt",
249
+ ]
250
+ if self.raw_prefix == "geom_drugs_revisited":
251
+ return [
252
+ f"data_geom_drugs_revisited_probmanifold_conformer_unit{suffix}.pt",
253
+ f"split_geom_drugs_revisited_probmanifold_conformer_unit{suffix}.pt",
254
+ ]
255
+ if self.raw_prefix.startswith("geom_drugs75k"):
256
+ return [
257
+ f"data_{self.raw_prefix}_probmanifold_conformer_unit{suffix}.pt",
258
+ f"split_{self.raw_prefix}_probmanifold_conformer_unit{suffix}.pt",
259
+ ]
260
+ return [
261
+ f"data_probmanifold_conformer_unit{suffix}.pt",
262
+ f"split_probmanifold_conformer_unit{suffix}.pt",
263
+ ]
264
+
265
+ def _resolve_max_confs(self, requested_max_confs):
266
+ """Resolve ``max_confs=0`` from train/validation conformer groups.
267
+
268
+ The conformer-ID embedding is trained only on train/validation data.
269
+ Test sets may contain many more reference conformers, but using their
270
+ maximum here would create embedding rows that never receive training
271
+ supervision. Full test ensembles are still available separately to
272
+ raw-reference evaluation in ``infer_base_conformers.py``.
273
+ """
274
+ if requested_max_confs > 0:
275
+ return requested_max_confs
276
+ if requested_max_confs < 0:
277
+ raise ValueError("max_confs must be positive, or 0 for automatic detection.")
278
+
279
+ # Raw GEOM files can contain RDKit objects serialized by a newer
280
+ # version. Their repetitive compatibility warnings add no diagnostic
281
+ # value during a count-only pass.
282
+ RDLogger.DisableLog("rdApp.warning")
283
+ maximum = 0
284
+ for filename in self.raw_file_names[:2]:
285
+ path = os.path.join(self.root, filename)
286
+ if not os.path.exists(path):
287
+ raise FileNotFoundError(path)
288
+ data_list = torch.load(path, map_location="cpu", weights_only=False)
289
+ counts = {}
290
+ for datapoint in data_list:
291
+ try:
292
+ mol = datapoint["rdmol"]
293
+ smiles = str(datapoint["smiles"])
294
+ pos = torch.as_tensor(datapoint["pos"])
295
+ except (KeyError, TypeError, AttributeError):
296
+ continue
297
+ if mol is None or pos.dim() != 2 or pos.size(0) != mol.GetNumAtoms():
298
+ continue
299
+ key = (smiles, int(mol.GetNumAtoms()))
300
+ counts[key] = counts.get(key, 0) + 1
301
+ if counts:
302
+ maximum = max(maximum, max(counts.values()))
303
+ del data_list
304
+
305
+ if maximum < 1:
306
+ raise RuntimeError("Automatic max_confs detection found no valid conformers.")
307
+ print(
308
+ f"Auto-detected max_confs={maximum} from train/validation data "
309
+ f"for raw_prefix={self.raw_prefix}."
310
+ )
311
+ return maximum
312
+
313
+ def _make_mol_with_conformer(self, mol, coords):
314
+ mol_conf = Chem.Mol(mol)
315
+
316
+ try:
317
+ mol_conf.RemoveAllConformers()
318
+ except Exception:
319
+ pass
320
+
321
+ conf = Chem.Conformer(mol_conf.GetNumAtoms())
322
+
323
+ coords = coords.cpu()
324
+
325
+ for atom_idx in range(mol_conf.GetNumAtoms()):
326
+ x, y, z = coords[atom_idx].tolist()
327
+ conf.SetAtomPosition(
328
+ atom_idx,
329
+ Point3D(float(x), float(y), float(z)),
330
+ )
331
+
332
+ mol_conf.AddConformer(conf, assignId=True)
333
+
334
+ return mol_conf
335
+
336
+ def _group_by_molecule(self, datalist, molecule_limit=0):
337
+ groups = OrderedDict()
338
+
339
+ for idx, datapoint in enumerate(datalist):
340
+ smiles = datapoint.get("smiles", None)
341
+ mol = datapoint.get("rdmol", None)
342
+
343
+ if smiles is None or mol is None:
344
+ continue
345
+
346
+ key = (
347
+ smiles,
348
+ mol.GetNumAtoms(),
349
+ )
350
+
351
+ if key not in groups:
352
+ if molecule_limit and len(groups) >= int(molecule_limit):
353
+ # Raw GEOM records are ordered by molecule. Once the
354
+ # requested prefix is complete, avoid retaining or
355
+ # processing the rest of a large debug split.
356
+ continue
357
+ groups[key] = []
358
+
359
+ groups[key].append((idx, datapoint))
360
+
361
+ return groups
362
+
363
+ def _process_split(
364
+ self,
365
+ datalist,
366
+ split_name,
367
+ mol_offset,
368
+ sample_offset,
369
+ molecule_limit=0,
370
+ ):
371
+ processed = []
372
+
373
+ groups = self._group_by_molecule(datalist, molecule_limit)
374
+
375
+ print(
376
+ f"[{split_name}] "
377
+ f"molecules={len(groups)} "
378
+ f"raw_conformers={len(datalist)}"
379
+ )
380
+
381
+ local_sample_id = 0
382
+
383
+ for local_mol_id, ((smiles, num_atoms), items) in enumerate(
384
+ tqdm(groups.items())
385
+ ):
386
+ global_mol_id = mol_offset + local_mol_id
387
+
388
+ mol = items[0][1]["rdmol"]
389
+
390
+ try:
391
+ x, edge_index, edge_attr, vdw_radii = smile2graph4GEOM(
392
+ items[0][1]
393
+ )
394
+ except Exception as e:
395
+ print(f"skip graph fail: {smiles}, {e}")
396
+ continue
397
+
398
+ if x.size(0) != num_atoms:
399
+ continue
400
+
401
+ edge_attr = enrich_edge_attr_with_bond_flags(
402
+ mol,
403
+ edge_index,
404
+ edge_attr,
405
+ )
406
+
407
+ torsion_index = build_torsion_index(mol)
408
+
409
+ valid = []
410
+
411
+ for conf_id, (data_id, datapoint) in enumerate(items):
412
+ # Full test references are an opt-in dataset mode for new
413
+ # ensemble studies; the historical protocol remains intact.
414
+ if (split_name != "test" or not self.keep_full_test_conformers) and len(valid) >= self.max_confs:
415
+ break
416
+
417
+ if "pos" not in datapoint:
418
+ continue
419
+
420
+ pos = torch.as_tensor(
421
+ datapoint["pos"],
422
+ dtype=torch.float,
423
+ )
424
+
425
+ if pos.dim() != 2:
426
+ continue
427
+
428
+ if pos.size(0) != x.size(0):
429
+ continue
430
+
431
+ valid.append(
432
+ (
433
+ conf_id,
434
+ data_id,
435
+ datapoint,
436
+ pos,
437
+ )
438
+ )
439
+
440
+ if len(valid) == 0:
441
+ continue
442
+
443
+ num_confs = len(valid)
444
+
445
+ for conf_id, data_id, datapoint, pos in valid:
446
+ sample_id = sample_offset + local_sample_id
447
+
448
+ mol_embedded = (self._make_mol_with_conformer(mol, pos)
449
+ if self.store_rdkit_objects else None)
450
+
451
+ data = Data(
452
+ x=x.clone(),
453
+ edge_index=edge_index.clone(),
454
+ edge_attr=edge_attr.clone(),
455
+ torsion_index=torsion_index.clone(),
456
+
457
+ pos=pos,
458
+
459
+ smiles=smiles,
460
+ mol_key=f"{smiles}__{num_atoms}",
461
+
462
+ mol_id=torch.tensor(
463
+ global_mol_id,
464
+ dtype=torch.long,
465
+ ),
466
+
467
+ split_mol_id=torch.tensor(
468
+ local_mol_id,
469
+ dtype=torch.long,
470
+ ),
471
+
472
+ conf_id=torch.tensor(
473
+ conf_id,
474
+ dtype=torch.long,
475
+ ),
476
+
477
+ idx=torch.tensor(
478
+ sample_id,
479
+ dtype=torch.long,
480
+ ),
481
+
482
+ sample_id=torch.tensor(
483
+ sample_id,
484
+ dtype=torch.long,
485
+ ),
486
+
487
+ num_atoms=torch.tensor(
488
+ x.size(0),
489
+ dtype=torch.long,
490
+ ),
491
+
492
+ num_confs_of_mol=torch.tensor(
493
+ num_confs,
494
+ dtype=torch.long,
495
+ ),
496
+
497
+ boltzmannweight=torch.tensor(
498
+ [
499
+ float(
500
+ datapoint.get(
501
+ "boltzmannweight",
502
+ 1.0,
503
+ )
504
+ )
505
+ ],
506
+ dtype=torch.float,
507
+ ),
508
+
509
+ totalenergy=torch.tensor(
510
+ [
511
+ float(
512
+ datapoint.get(
513
+ "totalenergy",
514
+ 0.0,
515
+ )
516
+ )
517
+ ],
518
+ dtype=torch.float,
519
+ ),
520
+
521
+ rdmol=mol if self.store_rdkit_objects else None,
522
+ rdmol_embedded=mol_embedded,
523
+ vdw_radii=vdw_radii,
524
+ )
525
+
526
+ data.batch_num_nodes = data.num_nodes
527
+
528
+ if self.pre_filter is not None:
529
+ if not self.pre_filter(data):
530
+ continue
531
+
532
+ if self.pre_transform is not None:
533
+ data = self.pre_transform(data)
534
+
535
+ processed.append(data)
536
+ local_sample_id += 1
537
+
538
+ return processed, len(processed), len(groups)
539
+
540
+ def process(self):
541
+ train_file, valid_file, test_file = self.raw_file_names
542
+
543
+ train_path = os.path.join(self.root, train_file)
544
+ valid_path = os.path.join(self.root, valid_file)
545
+ test_path = os.path.join(self.root, test_file)
546
+
547
+ if not os.path.exists(train_path):
548
+ raise FileNotFoundError(train_path)
549
+
550
+ if not os.path.exists(valid_path):
551
+ raise FileNotFoundError(valid_path)
552
+
553
+ if not os.path.exists(test_path):
554
+ raise FileNotFoundError(test_path)
555
+
556
+ # Load and release each raw split sequentially. This is important for
557
+ # large GEOM files: holding train, validation, and test RDKit objects
558
+ # simultaneously can exceed laptop memory before graph processing starts.
559
+ if self.process_train_mol_limit < 0:
560
+ train_list, train_num, train_mols = [], 0, 0
561
+ else:
562
+ train_data = torch.load(train_path, weights_only=False)
563
+ train_list, train_num, train_mols = self._process_split(
564
+ train_data,
565
+ "train",
566
+ 0,
567
+ 0,
568
+ self.process_train_mol_limit,
569
+ )
570
+ del train_data
571
+
572
+ if self.process_val_mol_limit < 0:
573
+ valid_list, valid_num, valid_mols = [], 0, 0
574
+ else:
575
+ valid_data = torch.load(valid_path, weights_only=False)
576
+ valid_list, valid_num, valid_mols = self._process_split(
577
+ valid_data,
578
+ "valid",
579
+ train_mols,
580
+ train_num,
581
+ self.process_val_mol_limit,
582
+ )
583
+ del valid_data
584
+
585
+ if self.process_test_mol_limit < 0:
586
+ # Training-only pilot caches do not need any test graphs. Avoid
587
+ # loading a large held-out raw file until the separate evaluation
588
+ # task requests it explicitly.
589
+ test_list, test_num, test_mols = [], 0, 0
590
+ else:
591
+ test_data = torch.load(test_path, weights_only=False)
592
+ test_list, test_num, test_mols = self._process_split(
593
+ test_data,
594
+ "test",
595
+ train_mols + valid_mols,
596
+ train_num + valid_num,
597
+ self.process_test_mol_limit,
598
+ )
599
+ del test_data
600
+
601
+ all_data = train_list + valid_list + test_list
602
+
603
+ train_index = list(range(train_num))
604
+
605
+ valid_index = list(
606
+ range(
607
+ train_num,
608
+ train_num + valid_num,
609
+ )
610
+ )
611
+
612
+ test_index = list(
613
+ range(
614
+ train_num + valid_num,
615
+ train_num + valid_num + test_num,
616
+ )
617
+ )
618
+
619
+ print(
620
+ f"train conformers={train_num}, "
621
+ f"valid={valid_num}, "
622
+ f"test={test_num}"
623
+ )
624
+
625
+ torch.save(
626
+ self.collate(all_data),
627
+ self.processed_paths[0],
628
+ )
629
+
630
+ with open(
631
+ self.processed_paths[1],
632
+ "wb",
633
+ ) as f:
634
+ pickle.dump(
635
+ [
636
+ train_index,
637
+ valid_index,
638
+ test_index,
639
+ ],
640
+ f,
641
+ )
642
+
643
+ def __repr__(self):
644
+ return f"{self.name}({len(self)})"