dualmapref 0.2.8__tar.gz
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.
- dualmapref-0.2.8/CITATION.cff +15 -0
- dualmapref-0.2.8/LICENSE +21 -0
- dualmapref-0.2.8/MANIFEST.in +14 -0
- dualmapref-0.2.8/PKG-INFO +167 -0
- dualmapref-0.2.8/README.md +136 -0
- dualmapref-0.2.8/checkpoints/README.md +7 -0
- dualmapref-0.2.8/configs/etflow_protocol_audit.yaml +64 -0
- dualmapref-0.2.8/configs/external_validation_development_matrix_v1.yaml +50 -0
- dualmapref-0.2.8/configs/external_validation_v1.yaml +85 -0
- dualmapref-0.2.8/configs/geomanimol.yaml +55 -0
- dualmapref-0.2.8/configs/hif2a_manimol_case_studies_clean.yaml +9 -0
- dualmapref-0.2.8/configs/manimol.yaml +4294 -0
- dualmapref-0.2.8/configs/mean_dispersion_denoising.yaml +20 -0
- dualmapref-0.2.8/configs/mean_dispersion_fixed_target.yaml +37 -0
- dualmapref-0.2.8/configs/mean_dispersion_local.yaml +125 -0
- dualmapref-0.2.8/configs/mixmanifold_server.yaml +284 -0
- dualmapref-0.2.8/configs/mixture_form.yaml +3484 -0
- dualmapref-0.2.8/docs/assets/conformend-figure1.png +0 -0
- dualmapref-0.2.8/docs/release.md +12 -0
- dualmapref-0.2.8/examples/basic_generation.py +18 -0
- dualmapref-0.2.8/examples/evaluate_conformers.py +19 -0
- dualmapref-0.2.8/examples/refine_candidates.py +66 -0
- dualmapref-0.2.8/examples/reproduce_route2_paper.sh +4 -0
- dualmapref-0.2.8/examples/run_mean_dispersion_denoising.sh +20 -0
- dualmapref-0.2.8/pyproject.toml +70 -0
- dualmapref-0.2.8/setup.cfg +4 -0
- dualmapref-0.2.8/src/args_parse.py +593 -0
- dualmapref-0.2.8/src/candidate_corrector.py +161 -0
- dualmapref-0.2.8/src/candidate_scorer.py +359 -0
- dualmapref-0.2.8/src/conformend/__init__.py +36 -0
- dualmapref-0.2.8/src/conformend/refiner.py +25 -0
- dualmapref-0.2.8/src/dataset/__init__.py +1 -0
- dualmapref-0.2.8/src/dataset/drugdataset.py +644 -0
- dualmapref-0.2.8/src/dataset/manifold.py +237 -0
- dualmapref-0.2.8/src/dataset/sharded_drugdataset.py +124 -0
- dualmapref-0.2.8/src/dataset/smiles2graph.py +217 -0
- dualmapref-0.2.8/src/dualmapref/__init__.py +24 -0
- dualmapref-0.2.8/src/dualmapref/refiner.py +16 -0
- dualmapref-0.2.8/src/dualmapref.egg-info/PKG-INFO +167 -0
- dualmapref-0.2.8/src/dualmapref.egg-info/SOURCES.txt +232 -0
- dualmapref-0.2.8/src/dualmapref.egg-info/dependency_links.txt +1 -0
- dualmapref-0.2.8/src/dualmapref.egg-info/entry_points.txt +10 -0
- dualmapref-0.2.8/src/dualmapref.egg-info/requires.txt +21 -0
- dualmapref-0.2.8/src/dualmapref.egg-info/top_level.txt +29 -0
- dualmapref-0.2.8/src/exputils.py +289 -0
- dualmapref-0.2.8/src/infer.py +23 -0
- dualmapref-0.2.8/src/infer_base_conformers.py +5589 -0
- dualmapref-0.2.8/src/infer_mean_dispersion_torsion.py +1117 -0
- dualmapref-0.2.8/src/infer_torsion.py +752 -0
- dualmapref-0.2.8/src/infer_torsional_denoiser.py +1021 -0
- dualmapref-0.2.8/src/manimol/__init__.py +30 -0
- dualmapref-0.2.8/src/manimol/__main__.py +6 -0
- dualmapref-0.2.8/src/manimol/cli.py +36 -0
- dualmapref-0.2.8/src/manimol/full_cli.py +41 -0
- dualmapref-0.2.8/src/manimol/legacy_imports.py +43 -0
- dualmapref-0.2.8/src/manimol/manifest.py +56 -0
- dualmapref-0.2.8/src/manimol/postprocess.py +148 -0
- dualmapref-0.2.8/src/manimol/priors.py +60 -0
- dualmapref-0.2.8/src/manimol/protocols.py +119 -0
- dualmapref-0.2.8/src/manimol/refiner.py +645 -0
- dualmapref-0.2.8/src/manimol/runtime.py +194 -0
- dualmapref-0.2.8/src/manimol/selection.py +85 -0
- dualmapref-0.2.8/src/manimol/torsion.py +22 -0
- dualmapref-0.2.8/src/manimol/train_cli.py +14 -0
- dualmapref-0.2.8/src/mixture_flow/src/__init__.py +1 -0
- dualmapref-0.2.8/src/mixture_flow/src/_path.py +7 -0
- dualmapref-0.2.8/src/mixture_flow/src/infer28_mixture_torsion_flow.py +1093 -0
- dualmapref-0.2.8/src/mixture_flow/src/mixture_torsion_flow.py +291 -0
- dualmapref-0.2.8/src/mixture_flow/src/summarize_mixture_debug.py +80 -0
- dualmapref-0.2.8/src/mixture_flow/src/train28_mixture_torsion_flow.py +583 -0
- dualmapref-0.2.8/src/mixture_torsion_flow.py +7 -0
- dualmapref-0.2.8/src/models/ manifold_learning.py +86 -0
- dualmapref-0.2.8/src/models/__init__.py +8 -0
- dualmapref-0.2.8/src/models/dist.py +273 -0
- dualmapref-0.2.8/src/models/dist2coords.py +23 -0
- dualmapref-0.2.8/src/models/egnn.py +107 -0
- dualmapref-0.2.8/src/models/experimental_backbones.py +364 -0
- dualmapref-0.2.8/src/models/gnnconv.py +512 -0
- dualmapref-0.2.8/src/models/kernel_inversion.py +141 -0
- dualmapref-0.2.8/src/models/kernels.py +427 -0
- dualmapref-0.2.8/src/models/losses.py +154 -0
- dualmapref-0.2.8/src/models/mean_dispersion_bridge.py +119 -0
- dualmapref-0.2.8/src/models/mean_dispersion_torsion_context.py +155 -0
- dualmapref-0.2.8/src/models/model.py +73 -0
- dualmapref-0.2.8/src/models/model0.py +2375 -0
- dualmapref-0.2.8/src/models/vis.py +128 -0
- dualmapref-0.2.8/src/pgraph_controls.py +107 -0
- dualmapref-0.2.8/src/scripts/__init__.py +2 -0
- dualmapref-0.2.8/src/scripts/analyze_basin_alignment.py +63 -0
- dualmapref-0.2.8/src/scripts/analyze_hif2a_conformer_aggregation.py +267 -0
- dualmapref-0.2.8/src/scripts/analyze_hif2a_contact_pharmacophore.py +1111 -0
- dualmapref-0.2.8/src/scripts/analyze_hif2a_fixed_budget.py +202 -0
- dualmapref-0.2.8/src/scripts/analyze_hif2a_pharmacophore.py +405 -0
- dualmapref-0.2.8/src/scripts/analyze_hif2a_pharmacophore_match_metric.py +386 -0
- dualmapref-0.2.8/src/scripts/analyze_p_control_candidate_flow.py +154 -0
- dualmapref-0.2.8/src/scripts/analyze_scaffold_basin.py +83 -0
- dualmapref-0.2.8/src/scripts/audit_dispersion_proposals.py +208 -0
- dualmapref-0.2.8/src/scripts/audit_etflow_protocol.py +258 -0
- dualmapref-0.2.8/src/scripts/audit_external_overlap.py +150 -0
- dualmapref-0.2.8/src/scripts/audit_oldbest_fulltrainval_split.py +82 -0
- dualmapref-0.2.8/src/scripts/audit_pdbbind_contact_pharmacophores.py +146 -0
- dualmapref-0.2.8/src/scripts/audit_qm9_geomol_test_split.py +94 -0
- dualmapref-0.2.8/src/scripts/audit_stage1_stage2_cross.py +131 -0
- dualmapref-0.2.8/src/scripts/build_geom_drugs_lowref_diagnostic.py +84 -0
- dualmapref-0.2.8/src/scripts/build_qm9_backfill_manifest.py +99 -0
- dualmapref-0.2.8/src/scripts/check_etflow_runtime.py +56 -0
- dualmapref-0.2.8/src/scripts/compare_prior_sources.py +122 -0
- dualmapref-0.2.8/src/scripts/compose_binding_mode_figure.py +87 -0
- dualmapref-0.2.8/src/scripts/compose_hif2a_contact_pharmacophore_suppfig.py +255 -0
- dualmapref-0.2.8/src/scripts/compose_hif2a_functional_geometry_pipeline.py +266 -0
- dualmapref-0.2.8/src/scripts/compose_manimol_internal_architecture.py +291 -0
- dualmapref-0.2.8/src/scripts/diagnose_flow_basin_intervention.py +54 -0
- dualmapref-0.2.8/src/scripts/diagnose_hif2a_pharmacophore_budget.py +185 -0
- dualmapref-0.2.8/src/scripts/diagnose_route2_p_ranking.py +123 -0
- dualmapref-0.2.8/src/scripts/discover_binding_mode_inputs.py +30 -0
- dualmapref-0.2.8/src/scripts/evaluate_functional_prior_stage1.py +230 -0
- dualmapref-0.2.8/src/scripts/evaluate_geomanimol.py +74 -0
- dualmapref-0.2.8/src/scripts/evaluate_geomanimol_flow.py +23 -0
- dualmapref-0.2.8/src/scripts/evaluate_geomanimol_multiscaffold.py +46 -0
- dualmapref-0.2.8/src/scripts/evaluate_geomanimol_residual.py +45 -0
- dualmapref-0.2.8/src/scripts/evaluate_geomanimol_torsion.py +40 -0
- dualmapref-0.2.8/src/scripts/evaluate_hif2a_bioactive_enrichment.py +91 -0
- dualmapref-0.2.8/src/scripts/evaluate_hif2a_pms_suite.py +440 -0
- dualmapref-0.2.8/src/scripts/evaluate_pdbbind_contact_pms.py +155 -0
- dualmapref-0.2.8/src/scripts/evaluate_platinum_external.py +114 -0
- dualmapref-0.2.8/src/scripts/evaluate_platinum_pool.py +82 -0
- dualmapref-0.2.8/src/scripts/evaluate_route2_calibrated_p.py +104 -0
- dualmapref-0.2.8/src/scripts/evaluate_route2_library_components.py +596 -0
- dualmapref-0.2.8/src/scripts/evaluate_route2_p_basin_budget.py +288 -0
- dualmapref-0.2.8/src/scripts/evaluate_stage1_checkpoint_only.py +30 -0
- dualmapref-0.2.8/src/scripts/export_dispersion_bond_audit.py +136 -0
- dualmapref-0.2.8/src/scripts/export_top20_shards.py +207 -0
- dualmapref-0.2.8/src/scripts/external_validation_metrics.py +152 -0
- dualmapref-0.2.8/src/scripts/fit_topology_prior.py +117 -0
- dualmapref-0.2.8/src/scripts/frozen_dispersion_audit.py +281 -0
- dualmapref-0.2.8/src/scripts/geomanimol_debug.py +50 -0
- dualmapref-0.2.8/src/scripts/geomanimol_smoke.py +37 -0
- dualmapref-0.2.8/src/scripts/infer_etflow_matched.py +524 -0
- dualmapref-0.2.8/src/scripts/inspect_checkpoint.py +12 -0
- dualmapref-0.2.8/src/scripts/inspect_conformer_capacity.py +71 -0
- dualmapref-0.2.8/src/scripts/inspect_pool.py +12 -0
- dualmapref-0.2.8/src/scripts/make_figure3_selection_loss.py +474 -0
- dualmapref-0.2.8/src/scripts/make_figure3_selection_summary.py +275 -0
- dualmapref-0.2.8/src/scripts/make_geom_debug_subset.py +62 -0
- dualmapref-0.2.8/src/scripts/make_low_reference_test_subset.py +70 -0
- dualmapref-0.2.8/src/scripts/merge_stage1_into_stage2.py +82 -0
- dualmapref-0.2.8/src/scripts/place_hif2a_candidates_for_evaluation.py +66 -0
- dualmapref-0.2.8/src/scripts/plot_hif2a_low_budget_efficiency.py +100 -0
- dualmapref-0.2.8/src/scripts/plot_prior_validation.py +48 -0
- dualmapref-0.2.8/src/scripts/precompute_stage2_assignments.py +90 -0
- dualmapref-0.2.8/src/scripts/prepare_bace_activity.py +98 -0
- dualmapref-0.2.8/src/scripts/prepare_bace_etkdg_benchmark.py +90 -0
- dualmapref-0.2.8/src/scripts/prepare_figure5_4x4_cases.py +470 -0
- dualmapref-0.2.8/src/scripts/prepare_geom_drugs_geomol_top20.py +472 -0
- dualmapref-0.2.8/src/scripts/prepare_pdbbind_core_raw.py +90 -0
- dualmapref-0.2.8/src/scripts/prepare_platinum_raw.py +74 -0
- dualmapref-0.2.8/src/scripts/prepare_pubchem_etkdg.py +54 -0
- dualmapref-0.2.8/src/scripts/rebuild_hif2a_final_sdf_and_metrics.py +55 -0
- dualmapref-0.2.8/src/scripts/rebuild_hif2a_fixed_pool_sweep.py +175 -0
- dualmapref-0.2.8/src/scripts/rebuild_hif2a_strict.py +220 -0
- dualmapref-0.2.8/src/scripts/recompute_hif2a_bootstrap.py +63 -0
- dualmapref-0.2.8/src/scripts/reference_free_geometry.py +112 -0
- dualmapref-0.2.8/src/scripts/render_binding_modes.py +413 -0
- dualmapref-0.2.8/src/scripts/render_figure5_4x4.py +330 -0
- dualmapref-0.2.8/src/scripts/render_figure6_pharmacophore_pymol.py +162 -0
- dualmapref-0.2.8/src/scripts/render_hif2a_contact_pharmacophore_pymol.py +414 -0
- dualmapref-0.2.8/src/scripts/render_hif2a_figure11.py +397 -0
- dualmapref-0.2.8/src/scripts/run_bace_2d_baseline.py +73 -0
- dualmapref-0.2.8/src/scripts/run_drugs_validation_matrix.py +225 -0
- dualmapref-0.2.8/src/scripts/run_external_reference_free.py +53 -0
- dualmapref-0.2.8/src/scripts/run_manimol_finalk_sensitivity.py +106 -0
- dualmapref-0.2.8/src/scripts/run_manimol_oversampling_sensitivity.py +109 -0
- dualmapref-0.2.8/src/scripts/run_pdisp_matched_intervention.py +116 -0
- dualmapref-0.2.8/src/scripts/run_stage1_stage2_cross20.py +83 -0
- dualmapref-0.2.8/src/scripts/score_hif2a_candidates_with_pgraph.py +221 -0
- dualmapref-0.2.8/src/scripts/summarize_dispersion_intervention.py +88 -0
- dualmapref-0.2.8/src/scripts/test_mean_dispersion_bridge.py +50 -0
- dualmapref-0.2.8/src/scripts/train_dispersion_head_diagnostic.py +320 -0
- dualmapref-0.2.8/src/scripts/train_dispersion_head_streaming.py +519 -0
- dualmapref-0.2.8/src/scripts/train_geomanimol.py +100 -0
- dualmapref-0.2.8/src/scripts/train_geomanimol_flow.py +26 -0
- dualmapref-0.2.8/src/scripts/train_geomanimol_residual.py +70 -0
- dualmapref-0.2.8/src/scripts/train_geomanimol_torsion.py +50 -0
- dualmapref-0.2.8/src/scripts/train_quantile_representation_diagnostic.py +204 -0
- dualmapref-0.2.8/src/scripts/validate_mean_dispersion_prior.py +173 -0
- dualmapref-0.2.8/src/scripts/validate_packed_shard.py +41 -0
- dualmapref-0.2.8/src/scripts/validate_stage1_prior.py +221 -0
- dualmapref-0.2.8/src/scripts/validate_top20_shards.py +36 -0
- dualmapref-0.2.8/src/scripts/validation/__init__.py +1 -0
- dualmapref-0.2.8/src/scripts/validation/hif2a_downstream_baselines.py +105 -0
- dualmapref-0.2.8/src/scripts/validation/hif2a_pairwise_ranker.py +218 -0
- dualmapref-0.2.8/src/scripts/verify_binding_mode_figure.py +47 -0
- dualmapref-0.2.8/src/torsion_manifold.py +281 -0
- dualmapref-0.2.8/src/train.py +78 -0
- dualmapref-0.2.8/src/train_graph_encoder.py +206 -0
- dualmapref-0.2.8/src/train_local_torsion_corrector.py +511 -0
- dualmapref-0.2.8/src/train_mean_dispersion_torsion_prior.py +329 -0
- dualmapref-0.2.8/src/train_probability_basin.py +235 -0
- dualmapref-0.2.8/src/train_torsion_generator.py +1631 -0
- dualmapref-0.2.8/src/train_torsion_manifold_prior.py +258 -0
- dualmapref-0.2.8/src/train_torsional_denoiser.py +415 -0
- dualmapref-0.2.8/src/trainer.py +1038 -0
- dualmapref-0.2.8/src/utils/__init__.py +7 -0
- dualmapref-0.2.8/src/utils/checkpoint.py +40 -0
- dualmapref-0.2.8/src/utils/device.py +17 -0
- dualmapref-0.2.8/src/utils/helpers.py +98 -0
- dualmapref-0.2.8/src/utils/kabsch.py +29 -0
- dualmapref-0.2.8/src/utils/lookup_table.py +164 -0
- dualmapref-0.2.8/src/utils/metrics.py +50 -0
- dualmapref-0.2.8/src/utils/optuna.py +111 -0
- dualmapref-0.2.8/src/utils/save_mol.py +366 -0
- dualmapref-0.2.8/src/utils/util.py +36 -0
- dualmapref-0.2.8/tests/conftest.py +21 -0
- dualmapref-0.2.8/tests/test_descriptive_names.py +46 -0
- dualmapref-0.2.8/tests/test_dispersion_head_diagnostic.py +49 -0
- dualmapref-0.2.8/tests/test_dispersion_head_streaming.py +82 -0
- dualmapref-0.2.8/tests/test_dispersion_intervention.py +45 -0
- dualmapref-0.2.8/tests/test_dispersion_torsion_context.py +54 -0
- dualmapref-0.2.8/tests/test_drugs_validation_matrix.py +72 -0
- dualmapref-0.2.8/tests/test_dualmapref_brand.py +11 -0
- dualmapref-0.2.8/tests/test_external_validation.py +144 -0
- dualmapref-0.2.8/tests/test_frozen_dispersion_audit.py +56 -0
- dualmapref-0.2.8/tests/test_infer_torsion_checkpoint_context.py +43 -0
- dualmapref-0.2.8/tests/test_legacy_imports.py +28 -0
- dualmapref-0.2.8/tests/test_manifest.py +20 -0
- dualmapref-0.2.8/tests/test_mean_dispersion_freezing.py +35 -0
- dualmapref-0.2.8/tests/test_oldbest_dispersion_context.py +76 -0
- dualmapref-0.2.8/tests/test_postprocess.py +32 -0
- dualmapref-0.2.8/tests/test_public_core.py +57 -0
- dualmapref-0.2.8/tests/test_public_protocol.py +68 -0
- dualmapref-0.2.8/tests/test_public_runtime.py +46 -0
- dualmapref-0.2.8/tests/test_refiner_module.py +132 -0
- dualmapref-0.2.8/tests/test_stage2_context.py +17 -0
- dualmapref-0.2.8/tests/test_trainer_progress.py +80 -0
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
cff-version: 1.2.0
|
|
2
|
+
title: "DualMapRef: Learning Ensemble Mean–Dispersion Pairwise Priors for Molecular Conformer Generation"
|
|
3
|
+
message: "If you use DualMapRef, please cite the associated manuscript."
|
|
4
|
+
type: software
|
|
5
|
+
authors:
|
|
6
|
+
- family-names: Nie
|
|
7
|
+
given-names: Jingzhe
|
|
8
|
+
- family-names: Li
|
|
9
|
+
given-names: Yong
|
|
10
|
+
- family-names: Zhang
|
|
11
|
+
given-names: Yanbing
|
|
12
|
+
- family-names: Shen
|
|
13
|
+
given-names: Wanxiang
|
|
14
|
+
version: 0.2.8
|
|
15
|
+
license: MIT
|
dualmapref-0.2.8/LICENSE
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 ConforMend developers
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
include README.md
|
|
2
|
+
include CITATION.cff
|
|
3
|
+
recursive-include src *.py
|
|
4
|
+
recursive-include configs *.yaml
|
|
5
|
+
recursive-include examples *.py *.sh
|
|
6
|
+
recursive-include tests *.py
|
|
7
|
+
recursive-include docs *.md
|
|
8
|
+
include docs/assets/conformend-figure1.png
|
|
9
|
+
recursive-include checkpoints README.md
|
|
10
|
+
prune data
|
|
11
|
+
prune dump
|
|
12
|
+
prune outputs
|
|
13
|
+
prune results
|
|
14
|
+
prune NCS_ManiMol
|
|
@@ -0,0 +1,167 @@
|
|
|
1
|
+
Metadata-Version: 2.1
|
|
2
|
+
Name: dualmapref
|
|
3
|
+
Version: 0.2.8
|
|
4
|
+
Summary: DualMapRef ensemble-relational conformer generation and compact-library construction
|
|
5
|
+
Author: DualMapRef developers
|
|
6
|
+
Keywords: molecular conformers,conformer generation,chemical informatics
|
|
7
|
+
Classifier: Development Status :: 3 - Alpha
|
|
8
|
+
Classifier: Intended Audience :: Science/Research
|
|
9
|
+
Classifier: Programming Language :: Python :: 3
|
|
10
|
+
Classifier: Topic :: Scientific/Engineering :: Chemistry
|
|
11
|
+
Requires-Python: >=3.9
|
|
12
|
+
Description-Content-Type: text/markdown
|
|
13
|
+
License-File: LICENSE
|
|
14
|
+
Requires-Dist: numpy>=1.23
|
|
15
|
+
Provides-Extra: full
|
|
16
|
+
Requires-Dist: torch>=2.3; extra == "full"
|
|
17
|
+
Requires-Dist: torch-geometric>=2.5; extra == "full"
|
|
18
|
+
Requires-Dist: rdkit; extra == "full"
|
|
19
|
+
Requires-Dist: scipy>=1.10; extra == "full"
|
|
20
|
+
Requires-Dist: tqdm>=4.66; extra == "full"
|
|
21
|
+
Requires-Dist: pyyaml>=6; extra == "full"
|
|
22
|
+
Provides-Extra: evaluation
|
|
23
|
+
Requires-Dist: pandas>=2.0; extra == "evaluation"
|
|
24
|
+
Requires-Dist: matplotlib>=3.7; extra == "evaluation"
|
|
25
|
+
Requires-Dist: scikit-learn>=1.3; extra == "evaluation"
|
|
26
|
+
Provides-Extra: test
|
|
27
|
+
Requires-Dist: pytest>=7; extra == "test"
|
|
28
|
+
Provides-Extra: build
|
|
29
|
+
Requires-Dist: build>=1.0; extra == "build"
|
|
30
|
+
Requires-Dist: twine>=4.0; extra == "build"
|
|
31
|
+
|
|
32
|
+
<p align="center">
|
|
33
|
+
<img src="docs/assets/dualmapref-logo.png" alt="DualMapRef" width="640">
|
|
34
|
+
</p>
|
|
35
|
+
|
|
36
|
+
<p align="center">
|
|
37
|
+
<a href="https://pypi.org/project/conformend/"><img src="https://img.shields.io/pypi/v/conformend" alt="PyPI"></a>
|
|
38
|
+
<a href="checkpoints/README.md"><img src="https://img.shields.io/badge/checkpoints-requirements-blue" alt="Checkpoints"></a>
|
|
39
|
+
<a href="examples/reproduce_route2_paper.sh"><img src="https://img.shields.io/badge/reproduction-script-527CC5" alt="Reproduction"></a>
|
|
40
|
+
<a href="CITATION.cff"><img src="https://img.shields.io/badge/citation-CFF-8755A5" alt="Citation"></a>
|
|
41
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-green" alt="MIT License"></a>
|
|
42
|
+
</p>
|
|
43
|
+
|
|
44
|
+
<p align="center">
|
|
45
|
+
<img src="https://img.shields.io/badge/arXiv-coming%20soon-lightgrey" alt="arXiv — coming soon">
|
|
46
|
+
<img src="https://img.shields.io/badge/Paper-coming%20soon-lightgrey" alt="Paper — coming soon">
|
|
47
|
+
<img src="https://img.shields.io/badge/DOI-coming%20soon-lightgrey" alt="DOI — coming soon">
|
|
48
|
+
</p>
|
|
49
|
+
|
|
50
|
+
# DualMapRef
|
|
51
|
+
|
|
52
|
+
DualMapRef generates compact molecular conformer ensembles using learned pairwise
|
|
53
|
+
mean and dispersion priors. The full pipeline starts from ETKDG structures,
|
|
54
|
+
denoises and refines the candidates, then selects a conformer library without
|
|
55
|
+
using reference conformers for selection.
|
|
56
|
+
|
|
57
|
+

|
|
58
|
+
|
|
59
|
+
## Install
|
|
60
|
+
|
|
61
|
+
Use Python 3.9 or newer. Install the current DualMapRef source from the repository root:
|
|
62
|
+
|
|
63
|
+
```bash
|
|
64
|
+
python -m pip install -e '.[full]'
|
|
65
|
+
```
|
|
66
|
+
|
|
67
|
+
The existing [PyPI package](https://pypi.org/project/conformend/) uses the former
|
|
68
|
+
name `conformend` and may differ from this source checkout. The `dualmapref`
|
|
69
|
+
package name below refers to this source checkout; publishing it on PyPI is a
|
|
70
|
+
separate release step.
|
|
71
|
+
|
|
72
|
+
For GPU inference, use a PyTorch build compatible with your CUDA environment.
|
|
73
|
+
|
|
74
|
+
## Select from existing conformers
|
|
75
|
+
|
|
76
|
+
```bash
|
|
77
|
+
dualmapref-select conformers.npy -k 5 -o selected_indices.npy
|
|
78
|
+
```
|
|
79
|
+
|
|
80
|
+
The input is a NumPy array of shape `[K, atoms, 3]` for one molecule with
|
|
81
|
+
consistent atom ordering. The output contains row indices into that array.
|
|
82
|
+
No model weights are required. Python users can call
|
|
83
|
+
`dualmapref.density_centrality_select` and `dualmapref.ensemble_pairwise_prior`.
|
|
84
|
+
|
|
85
|
+
## Run checkpoint-backed inference
|
|
86
|
+
|
|
87
|
+
**Reproduction status:** The `mean_dispersion_denoising` launcher below invokes the legacy
|
|
88
|
+
mean–dispersion / torsional-denoising pipeline. The standalone learned Refiner
|
|
89
|
+
is exposed separately in the next section; this launcher does not load its
|
|
90
|
+
checkpoint. Its equivalence to the complete serial DualMapRef workflow and the
|
|
91
|
+
paper’s reported benchmark results has not been established.
|
|
92
|
+
|
|
93
|
+
Supply the compatible base and torsional denoiser checkpoints, plus prepared
|
|
94
|
+
dataset records. The package and repository do not distribute these artifacts.
|
|
95
|
+
See [checkpoint requirements](checkpoints/README.md).
|
|
96
|
+
|
|
97
|
+
For `--raw_prefix geom_drugs`, the dataset loader expects `geom_drugs_train.pt`,
|
|
98
|
+
`geom_drugs_val.pt` and `geom_drugs_test.pt` under the data root when building
|
|
99
|
+
its cache. Use the archived data split for benchmark reproduction. The
|
|
100
|
+
preparation command below builds data using the GeoMol split:
|
|
101
|
+
|
|
102
|
+
```bash
|
|
103
|
+
python -m scripts.prepare_geom_drugs_geomol_top20 \
|
|
104
|
+
--workdir ./geom_work --output-dir ./data --output-prefix geom_drugs
|
|
105
|
+
```
|
|
106
|
+
|
|
107
|
+
```bash
|
|
108
|
+
dualmapref-infer \
|
|
109
|
+
--protocol mean_dispersion_denoising \
|
|
110
|
+
--base_checkpoint /path/to/stage2_base_rank0_best.pth \
|
|
111
|
+
--denoiser_checkpoint /path/to/torsion_denoiser_route2_oldbest_best.pth \
|
|
112
|
+
--dataset Drugs --data_root ./data --raw_prefix geom_drugs \
|
|
113
|
+
--split test --device cuda --output_dir results/mean_dispersion_denoising
|
|
114
|
+
```
|
|
115
|
+
|
|
116
|
+
The Python entry point is `DualMapRef.from_pretrained(...).generate(...)`.
|
|
117
|
+
The equivalent shell launcher is [run_mean_dispersion_denoising.sh](examples/run_mean_dispersion_denoising.sh); the existing [reproduce_route2_paper.sh](examples/reproduce_route2_paper.sh) remains a compatibility alias. The launcher reads `DUALMAPREF_CHECKPOINT`, `DUALMAPREF_DENOISER_CHECKPOINT`,
|
|
118
|
+
`DUALMAPREF_DATA_ROOT` and optionally `DUALMAPREF_DEVICE`. Earlier `CONFORMEND_*`
|
|
119
|
+
variables remain supported for compatibility.
|
|
120
|
+
|
|
121
|
+
## Refine candidates from another generator
|
|
122
|
+
|
|
123
|
+
Provide a single-molecule SDF template and an array of candidate coordinates in
|
|
124
|
+
ångströms. Use the same heavy atoms in the same order in both files. Supply a
|
|
125
|
+
compatible graph encoder and standalone Refiner checkpoint. This workflow
|
|
126
|
+
requires neither a denoiser checkpoint nor reference conformers.
|
|
127
|
+
|
|
128
|
+
```bash
|
|
129
|
+
python examples/refine_candidates.py \
|
|
130
|
+
--molecule molecule.sdf --coordinates conformers.npy \
|
|
131
|
+
--base-checkpoint /path/to/stage2_base_rank0_best.pth \
|
|
132
|
+
--refiner-checkpoint /path/to/strict_serial_refiner_localbasin_weighted_bestval_v4.pth \
|
|
133
|
+
--device cpu --output-dir results/refined
|
|
134
|
+
```
|
|
135
|
+
|
|
136
|
+
The example writes `refined_conformers.npy` in input order and `acceptance.json`
|
|
137
|
+
with a decision and reason for each candidate. Rejected updates retain their
|
|
138
|
+
input coordinates. The Python API is `dualmapref.refiner.DualMapRefRefiner`.
|
|
139
|
+
Support for external candidates describes the interface; measure refinement
|
|
140
|
+
quality on a paired benchmark for the generator you use.
|
|
141
|
+
|
|
142
|
+
## Source layout and compatibility
|
|
143
|
+
|
|
144
|
+
- `src/train_graph_encoder.py`: graph encoder training.
|
|
145
|
+
- `src/train_torsional_denoiser.py`: noise-conditioned torsional denoiser training.
|
|
146
|
+
- `src/manimol/refiner.py`: learned candidate-conditioned Refiner and inference API.
|
|
147
|
+
- `src/infer_mean_dispersion_torsion.py`: mean–dispersion and denoiser inference.
|
|
148
|
+
- `src/manimol/protocols.py`: named inference defaults; `route2_paper` is a legacy alias.
|
|
149
|
+
- `src/train_torsion_generator.py`, `src/train_probability_basin.py`, and
|
|
150
|
+
`src/train_local_torsion_corrector.py`: earlier experimental training paths.
|
|
151
|
+
- `src/infer_base_conformers.py` and `src/infer_torsional_denoiser.py`:
|
|
152
|
+
supporting inference implementations.
|
|
153
|
+
|
|
154
|
+
Historical numbered module names are mapped to their descriptive replacements by
|
|
155
|
+
`src/manimol/legacy_imports.py` after `manimol` is imported. Code loading a legacy
|
|
156
|
+
Python-serialized checkpoint directly should import `manimol` before `torch.load`.
|
|
157
|
+
The standalone Refiner is not automatically invoked by the denoising launcher.
|
|
158
|
+
|
|
159
|
+
## Development
|
|
160
|
+
|
|
161
|
+
```bash
|
|
162
|
+
python -m pip install -e '.[test,build]'
|
|
163
|
+
python -m pytest
|
|
164
|
+
python -m build
|
|
165
|
+
```
|
|
166
|
+
|
|
167
|
+
[Citation metadata](CITATION.cff) · [MIT License](LICENSE)
|
|
@@ -0,0 +1,136 @@
|
|
|
1
|
+
<p align="center">
|
|
2
|
+
<img src="docs/assets/dualmapref-logo.png" alt="DualMapRef" width="640">
|
|
3
|
+
</p>
|
|
4
|
+
|
|
5
|
+
<p align="center">
|
|
6
|
+
<a href="https://pypi.org/project/conformend/"><img src="https://img.shields.io/pypi/v/conformend" alt="PyPI"></a>
|
|
7
|
+
<a href="checkpoints/README.md"><img src="https://img.shields.io/badge/checkpoints-requirements-blue" alt="Checkpoints"></a>
|
|
8
|
+
<a href="examples/reproduce_route2_paper.sh"><img src="https://img.shields.io/badge/reproduction-script-527CC5" alt="Reproduction"></a>
|
|
9
|
+
<a href="CITATION.cff"><img src="https://img.shields.io/badge/citation-CFF-8755A5" alt="Citation"></a>
|
|
10
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-green" alt="MIT License"></a>
|
|
11
|
+
</p>
|
|
12
|
+
|
|
13
|
+
<p align="center">
|
|
14
|
+
<img src="https://img.shields.io/badge/arXiv-coming%20soon-lightgrey" alt="arXiv — coming soon">
|
|
15
|
+
<img src="https://img.shields.io/badge/Paper-coming%20soon-lightgrey" alt="Paper — coming soon">
|
|
16
|
+
<img src="https://img.shields.io/badge/DOI-coming%20soon-lightgrey" alt="DOI — coming soon">
|
|
17
|
+
</p>
|
|
18
|
+
|
|
19
|
+
# DualMapRef
|
|
20
|
+
|
|
21
|
+
DualMapRef generates compact molecular conformer ensembles using learned pairwise
|
|
22
|
+
mean and dispersion priors. The full pipeline starts from ETKDG structures,
|
|
23
|
+
denoises and refines the candidates, then selects a conformer library without
|
|
24
|
+
using reference conformers for selection.
|
|
25
|
+
|
|
26
|
+

|
|
27
|
+
|
|
28
|
+
## Install
|
|
29
|
+
|
|
30
|
+
Use Python 3.9 or newer. Install the current DualMapRef source from the repository root:
|
|
31
|
+
|
|
32
|
+
```bash
|
|
33
|
+
python -m pip install -e '.[full]'
|
|
34
|
+
```
|
|
35
|
+
|
|
36
|
+
The existing [PyPI package](https://pypi.org/project/conformend/) uses the former
|
|
37
|
+
name `conformend` and may differ from this source checkout. The `dualmapref`
|
|
38
|
+
package name below refers to this source checkout; publishing it on PyPI is a
|
|
39
|
+
separate release step.
|
|
40
|
+
|
|
41
|
+
For GPU inference, use a PyTorch build compatible with your CUDA environment.
|
|
42
|
+
|
|
43
|
+
## Select from existing conformers
|
|
44
|
+
|
|
45
|
+
```bash
|
|
46
|
+
dualmapref-select conformers.npy -k 5 -o selected_indices.npy
|
|
47
|
+
```
|
|
48
|
+
|
|
49
|
+
The input is a NumPy array of shape `[K, atoms, 3]` for one molecule with
|
|
50
|
+
consistent atom ordering. The output contains row indices into that array.
|
|
51
|
+
No model weights are required. Python users can call
|
|
52
|
+
`dualmapref.density_centrality_select` and `dualmapref.ensemble_pairwise_prior`.
|
|
53
|
+
|
|
54
|
+
## Run checkpoint-backed inference
|
|
55
|
+
|
|
56
|
+
**Reproduction status:** The `mean_dispersion_denoising` launcher below invokes the legacy
|
|
57
|
+
mean–dispersion / torsional-denoising pipeline. The standalone learned Refiner
|
|
58
|
+
is exposed separately in the next section; this launcher does not load its
|
|
59
|
+
checkpoint. Its equivalence to the complete serial DualMapRef workflow and the
|
|
60
|
+
paper’s reported benchmark results has not been established.
|
|
61
|
+
|
|
62
|
+
Supply the compatible base and torsional denoiser checkpoints, plus prepared
|
|
63
|
+
dataset records. The package and repository do not distribute these artifacts.
|
|
64
|
+
See [checkpoint requirements](checkpoints/README.md).
|
|
65
|
+
|
|
66
|
+
For `--raw_prefix geom_drugs`, the dataset loader expects `geom_drugs_train.pt`,
|
|
67
|
+
`geom_drugs_val.pt` and `geom_drugs_test.pt` under the data root when building
|
|
68
|
+
its cache. Use the archived data split for benchmark reproduction. The
|
|
69
|
+
preparation command below builds data using the GeoMol split:
|
|
70
|
+
|
|
71
|
+
```bash
|
|
72
|
+
python -m scripts.prepare_geom_drugs_geomol_top20 \
|
|
73
|
+
--workdir ./geom_work --output-dir ./data --output-prefix geom_drugs
|
|
74
|
+
```
|
|
75
|
+
|
|
76
|
+
```bash
|
|
77
|
+
dualmapref-infer \
|
|
78
|
+
--protocol mean_dispersion_denoising \
|
|
79
|
+
--base_checkpoint /path/to/stage2_base_rank0_best.pth \
|
|
80
|
+
--denoiser_checkpoint /path/to/torsion_denoiser_route2_oldbest_best.pth \
|
|
81
|
+
--dataset Drugs --data_root ./data --raw_prefix geom_drugs \
|
|
82
|
+
--split test --device cuda --output_dir results/mean_dispersion_denoising
|
|
83
|
+
```
|
|
84
|
+
|
|
85
|
+
The Python entry point is `DualMapRef.from_pretrained(...).generate(...)`.
|
|
86
|
+
The equivalent shell launcher is [run_mean_dispersion_denoising.sh](examples/run_mean_dispersion_denoising.sh); the existing [reproduce_route2_paper.sh](examples/reproduce_route2_paper.sh) remains a compatibility alias. The launcher reads `DUALMAPREF_CHECKPOINT`, `DUALMAPREF_DENOISER_CHECKPOINT`,
|
|
87
|
+
`DUALMAPREF_DATA_ROOT` and optionally `DUALMAPREF_DEVICE`. Earlier `CONFORMEND_*`
|
|
88
|
+
variables remain supported for compatibility.
|
|
89
|
+
|
|
90
|
+
## Refine candidates from another generator
|
|
91
|
+
|
|
92
|
+
Provide a single-molecule SDF template and an array of candidate coordinates in
|
|
93
|
+
ångströms. Use the same heavy atoms in the same order in both files. Supply a
|
|
94
|
+
compatible graph encoder and standalone Refiner checkpoint. This workflow
|
|
95
|
+
requires neither a denoiser checkpoint nor reference conformers.
|
|
96
|
+
|
|
97
|
+
```bash
|
|
98
|
+
python examples/refine_candidates.py \
|
|
99
|
+
--molecule molecule.sdf --coordinates conformers.npy \
|
|
100
|
+
--base-checkpoint /path/to/stage2_base_rank0_best.pth \
|
|
101
|
+
--refiner-checkpoint /path/to/strict_serial_refiner_localbasin_weighted_bestval_v4.pth \
|
|
102
|
+
--device cpu --output-dir results/refined
|
|
103
|
+
```
|
|
104
|
+
|
|
105
|
+
The example writes `refined_conformers.npy` in input order and `acceptance.json`
|
|
106
|
+
with a decision and reason for each candidate. Rejected updates retain their
|
|
107
|
+
input coordinates. The Python API is `dualmapref.refiner.DualMapRefRefiner`.
|
|
108
|
+
Support for external candidates describes the interface; measure refinement
|
|
109
|
+
quality on a paired benchmark for the generator you use.
|
|
110
|
+
|
|
111
|
+
## Source layout and compatibility
|
|
112
|
+
|
|
113
|
+
- `src/train_graph_encoder.py`: graph encoder training.
|
|
114
|
+
- `src/train_torsional_denoiser.py`: noise-conditioned torsional denoiser training.
|
|
115
|
+
- `src/manimol/refiner.py`: learned candidate-conditioned Refiner and inference API.
|
|
116
|
+
- `src/infer_mean_dispersion_torsion.py`: mean–dispersion and denoiser inference.
|
|
117
|
+
- `src/manimol/protocols.py`: named inference defaults; `route2_paper` is a legacy alias.
|
|
118
|
+
- `src/train_torsion_generator.py`, `src/train_probability_basin.py`, and
|
|
119
|
+
`src/train_local_torsion_corrector.py`: earlier experimental training paths.
|
|
120
|
+
- `src/infer_base_conformers.py` and `src/infer_torsional_denoiser.py`:
|
|
121
|
+
supporting inference implementations.
|
|
122
|
+
|
|
123
|
+
Historical numbered module names are mapped to their descriptive replacements by
|
|
124
|
+
`src/manimol/legacy_imports.py` after `manimol` is imported. Code loading a legacy
|
|
125
|
+
Python-serialized checkpoint directly should import `manimol` before `torch.load`.
|
|
126
|
+
The standalone Refiner is not automatically invoked by the denoising launcher.
|
|
127
|
+
|
|
128
|
+
## Development
|
|
129
|
+
|
|
130
|
+
```bash
|
|
131
|
+
python -m pip install -e '.[test,build]'
|
|
132
|
+
python -m pytest
|
|
133
|
+
python -m build
|
|
134
|
+
```
|
|
135
|
+
|
|
136
|
+
[Citation metadata](CITATION.cff) · [MIT License](LICENSE)
|
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
# Checkpoints
|
|
2
|
+
|
|
3
|
+
Model weights are intentionally not included in this source distribution.
|
|
4
|
+
Place a compatible checkpoint here locally, or pass an absolute path to the
|
|
5
|
+
inference commands. Checkpoint provenance, training configuration, and the
|
|
6
|
+
matching data split must be recorded with any released weight.
|
|
7
|
+
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
common:
|
|
2
|
+
python: /opt/anaconda3/envs/openff/bin/python
|
|
3
|
+
args:
|
|
4
|
+
data_root: data
|
|
5
|
+
raw_prefix: geom_drugs
|
|
6
|
+
split: test
|
|
7
|
+
dataset: Drugs
|
|
8
|
+
model: drugs-o3
|
|
9
|
+
reference_template: etflow
|
|
10
|
+
device: cuda
|
|
11
|
+
batch_size: 8
|
|
12
|
+
seed: 20260730
|
|
13
|
+
thresholds: 0.25,0.5,0.75,1.0,1.25,1.5,1.75,2.0
|
|
14
|
+
|
|
15
|
+
tasks:
|
|
16
|
+
preflight:
|
|
17
|
+
script: scripts/audit_etflow_protocol.py
|
|
18
|
+
args:
|
|
19
|
+
expected_test_molecules: 1000
|
|
20
|
+
output_prefix: outputs/etflow_audit/drugs_o3_preflight
|
|
21
|
+
|
|
22
|
+
official_2k_steps20_limit100:
|
|
23
|
+
script: scripts/infer_etflow_matched.py
|
|
24
|
+
args:
|
|
25
|
+
steps: 20
|
|
26
|
+
limit_molecules: 100
|
|
27
|
+
budget_mode: exact_2k
|
|
28
|
+
target_cap: 0
|
|
29
|
+
output_prefix: outputs/etflow_audit/drugs_o3_2k_steps20_limit100
|
|
30
|
+
|
|
31
|
+
official_2k_steps50_limit100:
|
|
32
|
+
script: scripts/infer_etflow_matched.py
|
|
33
|
+
args:
|
|
34
|
+
steps: 50
|
|
35
|
+
limit_molecules: 100
|
|
36
|
+
budget_mode: exact_2k
|
|
37
|
+
target_cap: 0
|
|
38
|
+
output_prefix: outputs/etflow_audit/drugs_o3_2k_steps50_limit100
|
|
39
|
+
|
|
40
|
+
official_2k_steps50_nochirality_limit100:
|
|
41
|
+
script: scripts/infer_etflow_matched.py
|
|
42
|
+
args:
|
|
43
|
+
steps: 50
|
|
44
|
+
limit_molecules: 100
|
|
45
|
+
budget_mode: exact_2k
|
|
46
|
+
target_cap: 0
|
|
47
|
+
disable_posthoc_chirality: true
|
|
48
|
+
output_prefix: outputs/etflow_audit/drugs_o3_2k_steps50_nochirality_limit100
|
|
49
|
+
|
|
50
|
+
rmsd_crosscheck10:
|
|
51
|
+
script: scripts/audit_etflow_protocol.py
|
|
52
|
+
args:
|
|
53
|
+
expected_test_molecules: 1000
|
|
54
|
+
crosscheck_molecules: 10
|
|
55
|
+
crosscheck_samples: 32
|
|
56
|
+
crosscheck_steps: 50
|
|
57
|
+
output_prefix: outputs/etflow_audit/drugs_o3_rmsd_crosscheck10
|
|
58
|
+
|
|
59
|
+
summarize_limit100:
|
|
60
|
+
script: scripts/audit_etflow_protocol.py
|
|
61
|
+
args:
|
|
62
|
+
expected_test_molecules: 1000
|
|
63
|
+
conditions: current_capped32=outputs/visualization/etflow_drugs_matched_limit50_cuda_curve.csv,official_2k_steps20=outputs/etflow_audit/drugs_o3_2k_steps20_limit100_curve.csv,official_2k_steps50=outputs/etflow_audit/drugs_o3_2k_steps50_limit100_curve.csv,official_2k_steps50_nochirality=outputs/etflow_audit/drugs_o3_2k_steps50_nochirality_limit100_curve.csv,full_official_protocol=outputs/etflow_audit/drugs_o3_2k_steps50_limit100_curve.csv
|
|
64
|
+
output_prefix: outputs/etflow_audit/drugs_o3_limit100_protocol
|
|
@@ -0,0 +1,50 @@
|
|
|
1
|
+
protocol_version: 1
|
|
2
|
+
status: frozen_before_full_external_evaluation
|
|
3
|
+
development_dataset: GEOM-Drugs-validation
|
|
4
|
+
external_datasets_forbidden_for_tuning:
|
|
5
|
+
- Platinum-Diverse
|
|
6
|
+
- PDBbind-v2020-refined
|
|
7
|
+
|
|
8
|
+
candidate_budget: 100
|
|
9
|
+
retained_budgets: [1, 5, 10, 20, 50]
|
|
10
|
+
guided_fractions: [0.25, 0.50, 0.75]
|
|
11
|
+
sampling_seeds: [20260911, 20260912, 20260913]
|
|
12
|
+
|
|
13
|
+
arms:
|
|
14
|
+
- name: etkdg_oversample_same_selector
|
|
15
|
+
unchanged_etkdg_fraction: 1.0
|
|
16
|
+
mean_guidance: false
|
|
17
|
+
dispersion_guidance: false
|
|
18
|
+
- name: manimol_mean
|
|
19
|
+
unchanged_etkdg_fraction: "1 - guided_fraction"
|
|
20
|
+
mean_guidance: true
|
|
21
|
+
dispersion_guidance: false
|
|
22
|
+
- name: manimol_mean_dispersion
|
|
23
|
+
unchanged_etkdg_fraction: "1 - guided_fraction"
|
|
24
|
+
mean_guidance: true
|
|
25
|
+
dispersion_guidance: true
|
|
26
|
+
|
|
27
|
+
selection:
|
|
28
|
+
method: density-centrality
|
|
29
|
+
alpha: 1.0
|
|
30
|
+
beta: 0.45
|
|
31
|
+
duplicate_radius_normalized: 0.04
|
|
32
|
+
|
|
33
|
+
selection_rule:
|
|
34
|
+
dataset: GEOM-Drugs-validation
|
|
35
|
+
budgets: [5, 20]
|
|
36
|
+
metrics:
|
|
37
|
+
maximize: [COV-R, COV-P]
|
|
38
|
+
minimize: [AMR-R, AMR-P]
|
|
39
|
+
note: >-
|
|
40
|
+
Select one operating point using the manuscript's balanced
|
|
41
|
+
recall-compactness interpretation. Do not use Platinum or PDBbind outcomes
|
|
42
|
+
to choose the allocation fraction, selector, threshold, or checkpoint.
|
|
43
|
+
|
|
44
|
+
external_reporting:
|
|
45
|
+
always_report:
|
|
46
|
+
- ETKDG-direct-K
|
|
47
|
+
- ETKDG-100-same-selector
|
|
48
|
+
- MANIMOL-mean
|
|
49
|
+
- MANIMOL-mean-dispersion
|
|
50
|
+
no_guaranteed_dominance_claim: true
|
|
@@ -0,0 +1,85 @@
|
|
|
1
|
+
protocol_version: external-validation-v1
|
|
2
|
+
frozen_on: 2026-09-12
|
|
3
|
+
|
|
4
|
+
scope:
|
|
5
|
+
model_training: frozen_geom_drugs_only
|
|
6
|
+
receptor_input_to_generator: false
|
|
7
|
+
crystal_input_to_generator_or_selector: false
|
|
8
|
+
claims:
|
|
9
|
+
- finite_budget_conformer_supply
|
|
10
|
+
- structural_diagnostics
|
|
11
|
+
excluded_claims:
|
|
12
|
+
- docking_superiority
|
|
13
|
+
- virtual_screening_superiority
|
|
14
|
+
- affinity_prediction
|
|
15
|
+
|
|
16
|
+
sampling:
|
|
17
|
+
seeds: [20260911, 20260912, 20260913]
|
|
18
|
+
retained_k: [1, 5, 10, 20, 50]
|
|
19
|
+
matched_internal_candidates: 100
|
|
20
|
+
selector:
|
|
21
|
+
name: density_centrality
|
|
22
|
+
density_neighbors: 20
|
|
23
|
+
alpha: 1.0
|
|
24
|
+
beta: 0.45
|
|
25
|
+
prefilter_quantile: 0.45
|
|
26
|
+
duplicate_radius_normalized: 0.04
|
|
27
|
+
|
|
28
|
+
platinum_diverse:
|
|
29
|
+
release: 2017-01
|
|
30
|
+
methods:
|
|
31
|
+
- etkdg_direct_k
|
|
32
|
+
- etkdg_oversample_density_centrality
|
|
33
|
+
- manimol_mean_only
|
|
34
|
+
- manimol_mean_dispersion
|
|
35
|
+
rmsd_thresholds_A: [0.5, 1.0, 1.5, 2.0]
|
|
36
|
+
primary_endpoints:
|
|
37
|
+
- recovery_1A_at_k5
|
|
38
|
+
- recovery_1A_at_k20
|
|
39
|
+
- recovery_2A_at_k5
|
|
40
|
+
- recovery_2A_at_k20
|
|
41
|
+
report_cov_p: false
|
|
42
|
+
|
|
43
|
+
pdbbind_refined:
|
|
44
|
+
release: v2020_refined
|
|
45
|
+
contact_profiler: plip
|
|
46
|
+
feature_definition: rdkit_BaseFeatures_fdef
|
|
47
|
+
feature_families: [donor, acceptor, aromatic, hydrophobe]
|
|
48
|
+
minimum_contact_defined_features: 2
|
|
49
|
+
region_sigma_A:
|
|
50
|
+
donor: 0.75
|
|
51
|
+
acceptor: 0.75
|
|
52
|
+
aromatic: 1.0
|
|
53
|
+
hydrophobe: 1.5
|
|
54
|
+
topology_scale_A: 1.5
|
|
55
|
+
pms_weights:
|
|
56
|
+
region: 0.5
|
|
57
|
+
topology: 0.5
|
|
58
|
+
pms_success_thresholds: [0.8, 0.9]
|
|
59
|
+
primary_endpoints:
|
|
60
|
+
- pms_success_0.8_at_k5
|
|
61
|
+
- pms_success_0.8_at_k20
|
|
62
|
+
|
|
63
|
+
overlap_audit:
|
|
64
|
+
exact_identifiers:
|
|
65
|
+
- canonical_isomeric_smiles
|
|
66
|
+
- standard_inchikey
|
|
67
|
+
exclude_exact_geom_drugs_train_identity: true
|
|
68
|
+
scaffold: bemis_murcko
|
|
69
|
+
fingerprint:
|
|
70
|
+
type: morgan
|
|
71
|
+
radius: 2
|
|
72
|
+
n_bits: 2048
|
|
73
|
+
nearest_train_tanimoto_bins: [0.0, 0.4, 0.6, 0.8, 1.000001]
|
|
74
|
+
remove_cross_benchmark_overlap: false
|
|
75
|
+
|
|
76
|
+
statistics:
|
|
77
|
+
primary_unit:
|
|
78
|
+
platinum: molecule
|
|
79
|
+
pdbbind: complex
|
|
80
|
+
hif2a: ligand
|
|
81
|
+
aggregate_seeds_within_unit: mean
|
|
82
|
+
paired_bootstrap_resamples: 10000
|
|
83
|
+
confidence_level: 0.95
|
|
84
|
+
pdbbind_cluster_sensitivity: uniprot_or_protein_family
|
|
85
|
+
generation_failure_counts_as_threshold_failure: true
|
|
@@ -0,0 +1,55 @@
|
|
|
1
|
+
common:
|
|
2
|
+
python: auto
|
|
3
|
+
args: {}
|
|
4
|
+
|
|
5
|
+
tasks:
|
|
6
|
+
train_flow_debug:
|
|
7
|
+
script: scripts/train_geomanimol_flow.py
|
|
8
|
+
args:
|
|
9
|
+
train: data/geom_drugs_geomol_top20/geom_drugs_geomol_top20_debug_train.pt
|
|
10
|
+
val: data/geom_drugs_geomol_top20/geom_drugs_geomol_top20_debug_val.pt
|
|
11
|
+
train-molecules: 20
|
|
12
|
+
val-molecules: 20
|
|
13
|
+
epochs: 20
|
|
14
|
+
output: outputs/geomanimol/flow_expert_debug_checkpoint.pt
|
|
15
|
+
|
|
16
|
+
diagnose_flow_basin_intervention:
|
|
17
|
+
script: scripts/diagnose_flow_basin_intervention.py
|
|
18
|
+
args:
|
|
19
|
+
checkpoint: outputs/geomanimol/flow_expert_debug_checkpoint.pt
|
|
20
|
+
molecules: 20
|
|
21
|
+
steps: 20
|
|
22
|
+
output: outputs/geomanimol/flow_basin_intervention.json
|
|
23
|
+
|
|
24
|
+
evaluate_flow_full:
|
|
25
|
+
script: scripts/evaluate_geomanimol_flow.py
|
|
26
|
+
args:
|
|
27
|
+
checkpoint: outputs/geomanimol/flow_expert_debug_checkpoint.pt
|
|
28
|
+
molecules: 20
|
|
29
|
+
scaffolds: 10
|
|
30
|
+
samples: 20
|
|
31
|
+
steps: 20
|
|
32
|
+
condition-mode: full
|
|
33
|
+
output: outputs/geomanimol/flow_full_steps20.json
|
|
34
|
+
|
|
35
|
+
evaluate_flow_no_z:
|
|
36
|
+
script: scripts/evaluate_geomanimol_flow.py
|
|
37
|
+
args:
|
|
38
|
+
checkpoint: outputs/geomanimol/flow_expert_debug_checkpoint.pt
|
|
39
|
+
molecules: 20
|
|
40
|
+
scaffolds: 10
|
|
41
|
+
samples: 20
|
|
42
|
+
steps: 20
|
|
43
|
+
condition-mode: no_z
|
|
44
|
+
output: outputs/geomanimol/flow_no_z_steps20.json
|
|
45
|
+
|
|
46
|
+
evaluate_flow_shuffle_z:
|
|
47
|
+
script: scripts/evaluate_geomanimol_flow.py
|
|
48
|
+
args:
|
|
49
|
+
checkpoint: outputs/geomanimol/flow_expert_debug_checkpoint.pt
|
|
50
|
+
molecules: 20
|
|
51
|
+
scaffolds: 10
|
|
52
|
+
samples: 20
|
|
53
|
+
steps: 20
|
|
54
|
+
condition-mode: shuffle_z
|
|
55
|
+
output: outputs/geomanimol/flow_shuffle_z_steps20.json
|