{ "model": "Boltz 2.2.1", "reference": "PDB 5T35 biological assembly 1", "scope": "Pre-cutoff retrospective minimal-versus-full system comparison", "protocol": { "seed": 42, "recycling_steps": 10, "diffusion_sampling_steps": 200, "samples_per_system": 5, "protein_msa": "MMseqs2", "ligand_representation": "canonical stereochemical SMILES for MZ1" }, "minimal_system": { "crane_run_id": "8971dc68-600f-4d1c-8216-9f99336852c4", "entities": [ "BRD4 BD2", "pVHL", "MZ1" ], "samples": [ { "sample": 0, "confidence_score": 0.9258, "BRD4_BD2_pVHL": { "dockq": 0.903, "irmsd_angstrom": 0.448, "lrmsd_angstrom": 2.436, "native_contact_fraction": 0.867 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 0.928, "after_pVHL_pocket_alignment": 0.642 } }, { "sample": 1, "confidence_score": 0.9225, "BRD4_BD2_pVHL": { "dockq": 0.921, "irmsd_angstrom": 0.391, "lrmsd_angstrom": 1.756, "native_contact_fraction": 0.867 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 0.745, "after_pVHL_pocket_alignment": 0.652 } }, { "sample": 2, "confidence_score": 0.9224, "BRD4_BD2_pVHL": { "dockq": 0.97, "irmsd_angstrom": 0.403, "lrmsd_angstrom": 1.315, "native_contact_fraction": 1.0 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 0.76, "after_pVHL_pocket_alignment": 0.608 } }, { "sample": 3, "confidence_score": 0.922, "BRD4_BD2_pVHL": { "dockq": 0.904, "irmsd_angstrom": 0.613, "lrmsd_angstrom": 2.466, "native_contact_fraction": 0.933 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 1.037, "after_pVHL_pocket_alignment": 0.721 } }, { "sample": 4, "confidence_score": 0.9219, "BRD4_BD2_pVHL": { "dockq": 0.941, "irmsd_angstrom": 0.457, "lrmsd_angstrom": 1.343, "native_contact_fraction": 0.933 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 0.941, "after_pVHL_pocket_alignment": 0.8 } } ] }, "full_system": { "crane_run_id": "637f0059-1aee-4808-aa70-ed25710ef9d3", "entities": [ "BRD4 BD2", "ElonginB", "ElonginC", "pVHL", "MZ1" ], "samples": [ { "sample": 0, "confidence_score": 0.9311, "BRD4_BD2_pVHL": { "dockq": 0.936, "irmsd_angstrom": 0.559, "lrmsd_angstrom": 2.32, "native_contact_fraction": 1.0 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 1.365, "after_pVHL_pocket_alignment": 0.982 } }, { "sample": 1, "confidence_score": 0.9307, "BRD4_BD2_pVHL": { "dockq": 0.94, "irmsd_angstrom": 0.453, "lrmsd_angstrom": 1.487, "native_contact_fraction": 0.933 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 0.896, "after_pVHL_pocket_alignment": 0.624 } }, { "sample": 2, "confidence_score": 0.9299, "BRD4_BD2_pVHL": { "dockq": 0.934, "irmsd_angstrom": 0.5, "lrmsd_angstrom": 1.496, "native_contact_fraction": 0.933 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 1.104, "after_pVHL_pocket_alignment": 0.759 } }, { "sample": 3, "confidence_score": 0.9298, "BRD4_BD2_pVHL": { "dockq": 0.922, "irmsd_angstrom": 0.56, "lrmsd_angstrom": 1.841, "native_contact_fraction": 0.933 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 0.978, "after_pVHL_pocket_alignment": 0.581 } }, { "sample": 4, "confidence_score": 0.9277, "BRD4_BD2_pVHL": { "dockq": 0.925, "irmsd_angstrom": 0.597, "lrmsd_angstrom": 1.277, "native_contact_fraction": 0.933 }, "MZ1_lrmsd_angstrom": { "after_BRD4_BD2_pocket_alignment": 1.085, "after_pVHL_pocket_alignment": 0.743 } } ] }, "summary": { "samples_recovering_high_quality_BRD4_BD2_pVHL_interface": "10 of 10", "dockq_range": [ 0.903, 0.97 ], "native_contacts_recovered_per_sample": "13 to 15 of 15", "MZ1_lrmsd_range_angstrom": [ 0.581, 1.365 ] }, "validation": { "dockq_rerun": { "date": "2026-07-28", "scope": "Full-system sample 1 only, the coordinate asset displayed in the story", "tool": "DockQ v2.1.3", "source_commit": "d9cbb1940bb0f42db3257f7da3b0e96f162b94d9", "mapping": "AD:AD model:native", "result_asset": "dockq-full-system-sample-1.json", "method_asset": "dockq-rerun.md", "model_sha256": "d749a3405b88b0b0ac4c287c48c9065dbd933d481c7fa228a406a6e7916112af", "native_sha256": "b702ef3bdc26045fd5f0c1bed7a23622a7493f79865b6ca6ea504ec118231b22", "result": { "dockq": 0.9400435090065002, "irmsd_angstrom": 0.45282110571861267, "lrmsd_angstrom": 1.4866536855697632, "native_contacts_recovered": "14 of 15" } } }, "methods": { "protein_interface": "DockQ v2.1.3 with fixed A:D to A:D chain mapping", "ligand_pose": "Symmetry-aware heavy-atom RMSD over 24 element-constrained distance-graph isomorphisms after fitting backbone atoms of reference-pocket residues within 10 angstrom of MZ1", "overlay_asset": "Full-system sample 1, selected retrospectively as the highest-DockQ full-system pose and globally fitted over 451 sequence-matched protein C-alpha atoms; global C-alpha RMSD 0.892 angstrom" }, "limitations": [ "PDB 5T35 was released on 2017-03-08, before Boltz-2's 2023-06-01 structural training-data cutoff, so it lies inside the model's structural training-data window. The report does not identify 5T35 as excluded, so it is not a documented holdout.", "This single case cannot diagnose overfitting or prove memorization.", "This reconstruction cannot establish generalization to unseen molecular systems.", "The controlled runs evaluated the ligand representation, MSA input, recycling depth, sampling count, and model version as one protocol package, so the outcome cannot be attributed to any one change.", "The five samples per system are correlated diffusion samples from one seed, not independent replicates.", "One molecular system does not establish general model performance.", "Structural reconstruction is not experimental evidence of binding cooperativity, ubiquitination geometry, or cellular degradation." ] }