[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75KCCMMMCWQKP8Z8377AX86":3,"skill-same:01M75KCCMMMCWQKP8Z8377AX86":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M75KCCMMMCWQKP8Z8377AX86","docking-rescoring",null,"Performs ML-based protein-ligand pose prediction and scoring using DiffDock-L (diffusion-based), Boltz-1 \u002F Boltz-2 (foundation model with affinity), Chai-1, AlphaFold3 ligand, EquiBind, TANKBind, NeuralPLexer, and hybrid workflows (DiffDock pose + GNINA rescore + PoseBusters QC). Explicit handling of when ML beats classical docking, when classical beats ML, the PB-invalid pose problem, and rescoring as the standard production hybrid. Use when modern docking is needed: foundation-model ligand-pose prediction, AI rescoring of classical poses, or scaffold-hopping in cross-docking scenarios.","pradyumnasagar\u002Fopen-research-skills","pradyumnasagar","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F10559871?v=4","01M75K78MYCCVKRTYGT8KKC380","Pradyumna Jayaram","01M75K78NTNE41F2PE1XBVFHAS",8,"MIT","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Fskills\u002Fchemoinformatics\u002Fdocking-rescoring\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\n\n\nname: docking-rescoring\ndescription: \"Performs ML-based protein-ligand pose prediction and scoring using DiffDock-L (diffusion-based), Boltz-1 \u002F Boltz-2 (foundation model with affinity), Chai-1, AlphaFold3 ligand, EquiBind, TANKBind, NeuralPLexer, and hybrid workflows (DiffDock pose + GNINA rescore + PoseBusters QC). Explicit handling of when ML beats classical docking, when classical beats ML, the PB-invalid pose problem, and rescoring as the standard production hybrid. Use when modern docking is needed: foundation-model ligand-pose prediction, AI rescoring of classical poses, or scaffold-hopping in cross-docking scenarios.\"\nlicense: MIT\n---\n\u003C!-- metadata:\n  category: chemoinformatics\n  oversight: low\n  version: 1.0.0\n  author: Pradyumna Jayaram\n  tags: [chemoinformatics, research]\n  difficulty: intermediate\n-->\n\n## Version Compatibility\n\nReference examples tested with: DiffDock-L (Corso 2024), Boltz-1 1.0+, Boltz-2 (Wohlwend 2025), Chai-1 0.4+, AlphaFold3 (DeepMind), EquiBind, TANKBind, GNINA 1.1+, PoseBusters 0.6+.\n\nBefore using code patterns, verify installed versions match. If versions differ:\n- Python: `pip show \u003Cpackage>` then `help(module.function)` to check signatures\n- CLI: `diffdock --version`; `boltz --version`\n\nIf code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.\n\n# ML Docking and Rescoring\n\n## Hard rules\n\n1. **No fabricated citations.** Every cited work must resolve to a verifiable\n2. **No claim without provenance.** Every quantitative or factual claim\n3. **No silent failure.** Every script invocation, API call, or tool use must declare its exit status and what to do on non-zero. A skill that silently swallows errors is a violation.\n\n## When to use\n\nLoad this skill when the user asks a question that matches its declared\ntrigger conditions (see the frontmatter `description` for the most common\nones). Do not load it for questions outside its scope — defer to the\nappropriate ORS skill instead.\n\n## When NOT to use\n\nDo not load this skill if the question is in a sibling skill's domain\n(see `## Cross-references`), if the user explicitly asks for a different\ntool, or if the task is outside the skill's declared category.\n\n## ML Docking Method Taxonomy\n\n| Tool | Approach | Speed | Strength | Fails when |\n|------|----------|-------|----------|------------|\n| DiffDock-L (Corso 2024) | Equivariant diffusion | 5s\u002Flig GPU | Pose sampling for cross-dock | ~50% PB-invalid; OOD |\n| Boltz-1 (Wohlwend 2024) | AlphaFold-style foundation | 10s GPU | Full complex prediction | DNA \u002F RNA may be off |\n| Boltz-2 (Wohlwend 2025) | Boltz-1 + affinity head | 10s GPU | Pose + affinity (Pearson 0.66 on 4-target FEP+ subset; RMSE ~1.5 kcal\u002Fmol on ChEMBL holdout) | Novel chemotype OOD |\n| Chai-1 (Chai 2024) | AlphaFold-style + LM | 10s GPU | Pose 77% RMSD success on PoseBusters | Limited public |\n| AlphaFold3 (DeepMind 2024) | Foundation model | API only | Pose 76% RMSD on PoseBusters | Restricted API access |\n| EquiBind | Equivariant single-shot | \u003C1s GPU | Fast pose | Lowest accuracy on PoseBusters |\n| TANKBind | Distance + classifier | \u003C1s GPU | Fast pose + score | Geometric inconsistency |\n| NeuralPLexer | E3-equivariant | \u003C1s | Fast pose | Limited adoption |\n| Glide (Schrödinger) | Hybrid grid + ML rescoring | 30s GPU | Commercial SOTA | License cost |\n| GNINA 1.1 CNN | Classical sampling + CNN scoring | 30s GPU | Best classical-hybrid | Limited to PDBbind chemotypes |\n\n**Decision:** For pose prediction with structure prediction needed, **Boltz-1** (or Boltz-2 if affinity also needed) is the modern open-source SOTA. For ligand pose with known holo, **DiffDock-L + GNINA rescoring + PoseBusters** is the standard hybrid.\n\n## Decision Tree by Scenario\n\n| Scenario | Recommended workflow |\n|----------|---------------------|\n| Known holo, need fast pose | GNINA classical |\n| Apo or AF-predicted protein, need pose | Boltz-1 or Chai-1 |\n| Cross-docking + scaffold hopping | DiffDock-L + GNINA rescore + PoseBusters |\n| Affinity prediction (replace FEP first-pass) | Boltz-2 affinity module |\n| Ultralarge library (1M+) | Vina pre-filter -> GNINA on top 1% -> Boltz-2 on top 0.1% |\n| Novel target family | Boltz-1 \u002F Chai-1 (uses MSA flexibility) |\n| Cofactor \u002F metal binding | AlphaFold3 (best cofactor handling); validate with classical |\n| PROTAC \u002F bivalent | Boltz-1 \u002F Chai-1 with multimer + constraints |\n| Production with auditable poses | GNINA classical + Boltz-2 score |\n\n## PoseBusters Problem (Critical)\n\nPoseBusters benchmark (Buttenschoen 2024) showed:\n\n| Tool | RMSD \u003C= 2 Å | PB-valid | RMSD \u003C= 2 Å AND PB-valid |\n|------|-------------|----------|--------------------------|\n| Vina (default) | 65% | 90% | 60% |\n| GOLD | 70% | 88% | 65% |\n| GNINA CNN | 73% | 85% | 65% |\n| DiffDock-L | 55% | 40% | 25% |\n| EquiBind | 30% | 25% | 10% |\n| TANKBind | 45% | 35% | 20% |\n| AlphaFold3 ligand | 76% | 65% | 55% |\n| Chai-1 | 77% | 70% | 58% |\n| Boltz-1 | 74% | 68% | 55% |\n| Boltz-2 (with affinity) | 76% | 70% | 58% |\n\n**Conclusion:** Modern foundation models match classical RMSD but with worse physical plausibility. Always require PB-valid + RMSD \u003C= 2 Å.\n\n## DiffDock-L + GNINA Hybrid Workflow (Production Standard)\n\n**Goal:** Use DiffDock-L for fast diverse pose sampling; GNINA CNN to rescore; PoseBusters to filter.\n\n```bash\n# Step 1: DiffDock-L pose sampling\npython -m inference \\\n    --protein_path receptor.pdb \\\n    --ligand_description smiles.smi \\\n    --out_dir diffdock_out\u002F \\\n    --samples_per_complex 40 \\\n    --inference_steps 20\n\n# Step 2: GNINA CNN rescoring\ngnina -r receptor.pdb -l diffdock_out\u002Ftop_poses.sdf \\\n      --autobox_ligand reference_ligand.sdf \\\n      --cnn_scoring rescore \\\n      -o gnina_rescored.sdf.gz\n\n# Step 3: PoseBusters QC\"\npython -c \"\nfrom posebusters import PoseBusters\nb = PoseBusters(config='dock')\ndf = b.bust(mol_pred='gnina_rescored.sdf.gz', mol_cond='receptor.pdb')\nprint(df.filter(regex='pass').sum().sum(), 'checks passed')\n\"\n```\n\n## Boltz-2 Affinity Prediction (Fast FEP Replacement)\n\n**Goal:** Predict binding affinity at ~10s\u002Fligand on GPU vs hours for FEP.\n\n**Approach:** Submit protein + ligand to Boltz-2 with the affinity head; receive predicted pIC50 or ΔG with uncertainty.\n\n```bash\nboltz predict input.yaml --use_affinity --output_dir boltz2_out\u002F\n```\n\n**Validation:** Boltz-2 reports Pearson 0.66 vs experimental on 4-target FEP+ subset; RMSE ~1.5 kcal\u002Fmol on ChEMBL holdout. Faster than FEP (10s vs hours) but ~3x less accurate.\n\n## Foundation Models: Boltz-1 and Chai-1\n\nFor novel target or no MSA, the foundation models treat ligand + protein jointly:\n\n```bash\n# Boltz-1\nboltz predict input.yaml --output_dir boltz1_out\u002F\n\n# Chai-1\nchai-lab predict input.fasta --output-dir chai_out\u002F\n```\n\n**Boltz-1 \u002F Chai-1 caveats:** Both accept MSA but treat ligand in pocket as constraints; output is the full complex structure. Ligand pose quality matches PoseBusters benchmarks in the table above.\n\n## Per-Tool Failure Modes\n\n### DiffDock-L -- PB-invalid poses\n\n**Trigger:** Default DiffDock-L output.\n\n**Mechanism:** Diffusion lacks explicit physical-plausibility term; ~50% of poses fail planarity, vdW overlap, or chirality tests.\n\n**Symptom:** Poses look reasonable in 2D depiction but fail QC.\n\n**Fix:** Always apply PoseBusters filter; rescore with GNINA.\n\n### Boltz-2 -- poor affinity for novel chemotype\n\n**Trigger:** Ligand chemotype outside ChEMBL training distribution.\n\n**Mechanism:** Affinity head trained on public bioactivity data; extrapolation is unreliable.\n\n**Symptom:** Predicted pIC50 far from experimental; uncertainty band very wide.\n\n**Fix:** Use Boltz-2 for first-pass triage only; FEP for top hits.\n\n### EquiBind -- worst accuracy\n\n**Trigger:** Default EquiBind without ensemble.\n\n**Mechanism:** Single-shot equivariant network; no refinement step.\n\n**Symptom:** Poses within RMSD > 5 Å.\n\n**Fix:** Avoid EquiBind; use DiffDock-L or Boltz-1 for sampling, then refine with GNINA.\n\n### AlphaFold3 -- restricted API\n\n**Trigger:** Academic users without commercial agreement.\n\n**Mechanism:** AF3 server requires Google Cloud account + quota.\n\n**Symptom:** Cannot submit jobs; rate-limited.\n\n**Fix:** Use Boltz-1 (open source) for similar performance; AF3 only for cross-validation.\n\n### ML methods ignore cryptic pockets\n\n**Trigger:** Binding site not in receptor conformation.\n\n**Mechanism:** ML methods score poses against static receptor.\n\n**Symptom:** All poses report poor affinity; known active missed.\n\n**Fix:** Pre-generate receptor ensemble via MD or use AlphaFold3 holo prediction.\n\n## Reconciliation: ML vs Classical\n\n| Aspect | Classical (Vina\u002FGNINA) | ML (DiffDock\u002FBoltz) |\n|--------|------------------------|---------------------|\n| Pose RMSD | 60-70% within 2 Å | 50-77% within 2 Å |\n| PB-validity | 85-90% | 40-70% |\n| Affinity accuracy | Correlates ~0.5 with exp | Boltz-2 ~0.66 Pearson on benchmark |\n| Speed | 5-30s\u002Flig | 5-10s GPU |\n| Out-of-distribution | Robust for chemotypes | Worse for novel scaffolds |\n| Interpretability | Force-field based | Black-box |\n\n**Production hybrid:** ML for pose sampling (broader search), classical for affinity + physical validation.\n\n## Common Errors\n\n| Symptom | Cause | Fix |\n|---------|-------|-----|\n| Boltz-2 OOM on big protein | MSA + ligand fits in 16GB | Reduce MSA depth; use Boltz-1 |\n| DiffDock all poses cluster | Insufficient samples | Increase `samples_per_complex` to 40-100 |\n| Boltz output not a single chain | Tokenizer confused by modified residues | Strip non-standard residues; use UniProt canonical |\n| GNINA can't read DiffDock output | SDF missing 3D | DiffDock writes 3D SDF; verify with PyMOL |\n| Chai-1 ignores pocket | Wrong binding-site hint | Pass `--pocket-residues` if available |\n\n## References\n\n- Corso et al., *Nat. Mach. Intell.* -- DiffDock-L.\n- Wohlwend et al., 2024 -- Boltz-1; 2025 -- Boltz-2 with affinity.\n- Buttenschoen et al., *Chem. Sci.* 15:3130 -- PoseBusters benchmark.\n- McNutt et al., *J. Cheminformatics* 13:43 -- GNINA 1.0 CNN.\n- Krishna et al., 2024 -- AlphaFold3 server.\n\n## Related Skills\n\n- chemoinformatics\u002Fvirtual-screening - Classical docking\n- chemoinformatics\u002Fpose-validation - PoseBusters QC\n- chemoinformatics\u002Ffree-energy-calculations - Post-docking FEP\n- chemoinformatics\u002Fcovalent-design - Covalent docking\n- chemoinformatics\u002Fprotac-degraders - Ternary complex prediction\n\n## Cross-references\n\nOther skills in this category:\n\n- admet-prediction\n- conformer-generation\n- covalent-design\n- free-energy-calculations\n- generative-design\n- molecular-descriptors\n- molecular-io\n- molecular-standardization\n- pharmacophore-modeling\n- pose-validation\n- protac-degraders\n- qsar-modeling\n- retrosynthesis\n- scaffold-analysis\n- shape-similarity\n- similarity-searching\n- substructure-search\n- virtual-screening\n\n## Changelog\n\n- **1.1.0** (migration) — Bulk-migrated to v0.4.0 schema: canonical\n  metadata block, base Hard rules, Cross-references. Body content\n  unchanged; author should review and fill in any domain-specific\n  extensions to the Hard rules.\n- **1.0.0** — Initial release.\n\n","\u003C!-- metadata:\n  category: chemoinformatics\n  oversight: low\n  version: 1.0.0\n  author: Pradyumna Jayaram\n  tags: [chemoinformatics, research]\n  difficulty: intermediate\n-->\n\n## Version Compatibility\n\nReference examples tested with: DiffDock-L (Corso 2024), Boltz-1 1.0+, Boltz-2 (Wohlwend 2025), Chai-1 0.4+, AlphaFold3 (DeepMind), EquiBind, TANKBind, GNINA 1.1+, PoseBusters 0.6+.\n\nBefore using code patterns, verify installed versions match. If versions differ:\n- Python: `pip show \u003Cpackage>` then `help(module.function)` to check signatures\n- CLI: `diffdock --version`; `boltz --version`\n\nIf code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.\n\n# ML Docking and Rescoring\n\n## Hard rules\n\n1. **No fabricated citations.** Every cited work must resolve to a verifiable\n2. **No claim without provenance.** Every quantitative or factual claim\n3. **No silent failure.** Every script invocation, API call, or tool use must declare its exit status and what to do on non-zero. A skill that silently swallows errors is a violation.\n\n## When to use\n\nLoad this skill when the user asks a question that matches its declared\ntrigger conditions (see the frontmatter `description` for the most common\nones). Do not load it for questions outside its scope — defer to the\nappropriate ORS skill instead.\n\n## When NOT to use\n\nDo not load this skill if the question is in a sibling skill's domain\n(see `## Cross-references`), if the user explicitly asks for a different\ntool, or if the task is outside the skill's declared category.\n\n## ML Docking Method Taxonomy\n\n| Tool | Approach | Speed | Strength | Fails when |\n|------|----------|-------|----------|------------|\n| DiffDock-L (Corso 2024) | Equivariant diffusion | 5s\u002Flig GPU | Pose sampling for cross-dock | ~50% PB-invalid; OOD |\n| Boltz-1 (Wohlwend 2024) | AlphaFold-style foundation | 10s GPU | Full complex prediction | DNA \u002F RNA may be off |\n| Boltz-2 (Wohlwend 2025) | Boltz-1 + affinity head | 10s GPU | Pose + affinity (Pearson 0.66 on 4-target FEP+ subset; RMSE ~1.5 kcal\u002Fmol on ChEMBL holdout) | Novel chemotype OOD |\n| Chai-1 (Chai 2024) | AlphaFold-style + LM | 10s GPU | Pose 77% RMSD success on PoseBusters | Limited public |\n| AlphaFold3 (DeepMind 2024) | Foundation model | API only | Pose 76% RMSD on PoseBusters | Restricted API access |\n| EquiBind | Equivariant single-shot | \u003C1s GPU | Fast pose | Lowest accuracy on PoseBusters |\n| TANKBind | Distance + classifier | \u003C1s GPU | Fast pose + score | Geometric inconsistency |\n| NeuralPLexer | E3-equivariant | \u003C1s | Fast pose | Limited adoption |\n| Glide (Schrödinger) | Hybrid grid + ML rescoring | 30s GPU | Commercial SOTA | License cost |\n| GNINA 1.1 CNN | Classical sampling + CNN scoring | 30s GPU | Best classical-hybrid | Limited to PDBbind chemotypes |\n\n**Decision:** For pose prediction with structure prediction needed, **Boltz-1** (or Boltz-2 if affinity also needed) is the modern open-source SOTA. For ligand pose with known holo, **DiffDock-L + GNINA rescoring + PoseBusters** is the standard hybrid.\n\n## Decision Tree by Scenario\n\n| Scenario | Recommended workflow |\n|----------|---------------------|\n| Known holo, need fast pose | GNINA classical |\n| Apo or AF-predicted protein, need pose | Boltz-1 or Chai-1 |\n| Cross-docking + scaffold hopping | DiffDock-L + GNINA rescore + PoseBusters |\n| Affinity prediction (replace FEP first-pass) | Boltz-2 affinity module |\n| Ultralarge library (1M+) | Vina pre-filter -> GNINA on top 1% -> Boltz-2 on top 0.1% |\n| Novel target family | Boltz-1 \u002F Chai-1 (uses MSA flexibility) |\n| Cofactor \u002F metal binding | AlphaFold3 (best cofactor handling); validate with classical |\n| PROTAC \u002F bivalent | Boltz-1 \u002F Chai-1 with multimer + constraints |\n| Production with auditable poses | GNINA classical + Boltz-2 score |\n\n## PoseBusters Problem (Critical)\n\nPoseBusters benchmark (Buttenschoen 2024) showed:\n\n| Tool | RMSD \u003C= 2 Å | PB-valid | RMSD \u003C= 2 Å AND PB-valid |\n|------|-------------|----------|--------------------------|\n| Vina (default) | 65% | 90% | 60% |\n| GOLD | 70% | 88% | 65% |\n| GNINA CNN | 73% | 85% | 65% |\n| DiffDock-L | 55% | 40% | 25% |\n| EquiBind | 30% | 25% | 10% |\n| TANKBind | 45% | 35% | 20% |\n| AlphaFold3 ligand | 76% | 65% | 55% |\n| Chai-1 | 77% | 70% | 58% |\n| Boltz-1 | 74% | 68% | 55% |\n| Boltz-2 (with affinity) | 76% | 70% | 58% |\n\n**Conclusion:** Modern foundation models match classical RMSD but with worse physical plausibility. Always require PB-valid + RMSD \u003C= 2 Å.\n\n## DiffDock-L + GNINA Hybrid Workflow (Production Standard)\n\n**Goal:** Use DiffDock-L for fast diverse pose sampling; GNINA CNN to rescore; PoseBusters to filter.\n\n```bash\n# Step 1: DiffDock-L pose sampling\npython -m inference \\\n    --protein_path receptor.pdb \\\n    --ligand_description smiles.smi \\\n    --out_dir diffdock_out\u002F \\\n    --samples_per_complex 40 \\\n    --inference_steps 20\n\n# Step 2: GNINA CNN rescoring\ngnina -r receptor.pdb -l diffdock_out\u002Ftop_poses.sdf \\\n      --autobox_ligand reference_ligand.sdf \\\n      --cnn_scoring rescore \\\n      -o gnina_rescored.sdf.gz\n\n# Step 3: PoseBusters QC\"\npython -c \"\nfrom posebusters import PoseBusters\nb = PoseBusters(config='dock')\ndf = b.bust(mol_pred='gnina_rescored.sdf.gz', mol_cond='receptor.pdb')\nprint(df.filter(regex='pass').sum().sum(), 'checks passed')\n\"\n```\n\n## Boltz-2 Affinity Prediction (Fast FEP Replacement)\n\n**Goal:** Predict binding affinity at ~10s\u002Fligand on GPU vs hours for FEP.\n\n**Approach:** Submit protein + ligand to Boltz-2 with the affinity head; receive predicted pIC50 or ΔG with uncertainty.\n\n```bash\nboltz predict input.yaml --use_affinity --output_dir boltz2_out\u002F\n```\n\n**Validation:** Boltz-2 reports Pearson 0.66 vs experimental on 4-target FEP+ subset; RMSE ~1.5 kcal\u002Fmol on ChEMBL holdout. Faster than FEP (10s vs hours) but ~3x less accurate.\n\n## Foundation Models: Boltz-1 and Chai-1\n\nFor novel target or no MSA, the foundation models treat ligand + protein jointly:\n\n```bash\n# Boltz-1\nboltz predict input.yaml --output_dir boltz1_out\u002F\n\n# Chai-1\nchai-lab predict input.fasta --output-dir chai_out\u002F\n```\n\n**Boltz-1 \u002F Chai-1 caveats:** Both accept MSA but treat ligand in pocket as constraints; output is the full complex structure. Ligand pose quality matches PoseBusters benchmarks in the table above.\n\n## Per-Tool Failure Modes\n\n### DiffDock-L -- PB-invalid poses\n\n**Trigger:** Default DiffDock-L output.\n\n**Mechanism:** Diffusion lacks explicit physical-plausibility term; ~50% of poses fail planarity, vdW overlap, or chirality tests.\n\n**Symptom:** Poses look reasonable in 2D depiction but fail QC.\n\n**Fix:** Always apply PoseBusters filter; rescore with GNINA.\n\n### Boltz-2 -- poor affinity for novel chemotype\n\n**Trigger:** Ligand chemotype outside ChEMBL training distribution.\n\n**Mechanism:** Affinity head trained on public bioactivity data; extrapolation is unreliable.\n\n**Symptom:** Predicted pIC50 far from experimental; uncertainty band very wide.\n\n**Fix:** Use Boltz-2 for first-pass triage only; FEP for top hits.\n\n### EquiBind -- worst accuracy\n\n**Trigger:** Default EquiBind without ensemble.\n\n**Mechanism:** Single-shot equivariant network; no refinement step.\n\n**Symptom:** Poses within RMSD > 5 Å.\n\n**Fix:** Avoid EquiBind; use DiffDock-L or Boltz-1 for sampling, then refine with GNINA.\n\n### AlphaFold3 -- restricted API\n\n**Trigger:** Academic users without commercial agreement.\n\n**Mechanism:** AF3 server requires Google Cloud account + quota.\n\n**Symptom:** Cannot submit jobs; rate-limited.\n\n**Fix:** Use Boltz-1 (open source) for similar performance; AF3 only for cross-validation.\n\n### ML methods ignore cryptic pockets\n\n**Trigger:** Binding site not in receptor conformation.\n\n**Mechanism:** ML methods score poses against static receptor.\n\n**Symptom:** All poses report poor affinity; known active missed.\n\n**Fix:** Pre-generate receptor ensemble via MD or use AlphaFold3 holo prediction.\n\n## Reconciliation: ML vs Classical\n\n| Aspect | Classical (Vina\u002FGNINA) | ML (DiffDock\u002FBoltz) |\n|--------|------------------------|---------------------|\n| Pose RMSD | 60-70% within 2 Å | 50-77% within 2 Å |\n| PB-validity | 85-90% | 40-70% |\n| Affinity accuracy | Correlates ~0.5 with exp | Boltz-2 ~0.66 Pearson on benchmark |\n| Speed | 5-30s\u002Flig | 5-10s GPU |\n| Out-of-distribution | Robust for chemotypes | Worse for novel scaffolds |\n| Interpretability | Force-field based | Black-box |\n\n**Production hybrid:** ML for pose sampling (broader search), classical for affinity + physical validation.\n\n## Common Errors\n\n| Symptom | Cause | Fix |\n|---------|-------|-----|\n| Boltz-2 OOM on big protein | MSA + ligand fits in 16GB | Reduce MSA depth; use Boltz-1 |\n| DiffDock all poses cluster | Insufficient samples | Increase `samples_per_complex` to 40-100 |\n| Boltz output not a single chain | Tokenizer confused by modified residues | Strip non-standard residues; use UniProt canonical |\n| GNINA can't read DiffDock output | SDF missing 3D | DiffDock writes 3D SDF; verify with PyMOL |\n| Chai-1 ignores pocket | Wrong binding-site hint | Pass `--pocket-residues` if available |\n\n## References\n\n- Corso et al., *Nat. Mach. Intell.* -- DiffDock-L.\n- Wohlwend et al., 2024 -- Boltz-1; 2025 -- Boltz-2 with affinity.\n- Buttenschoen et al., *Chem. Sci.* 15:3130 -- PoseBusters benchmark.\n- McNutt et al., *J. Cheminformatics* 13:43 -- GNINA 1.0 CNN.\n- Krishna et al., 2024 -- AlphaFold3 server.\n\n## Related Skills\n\n- chemoinformatics\u002Fvirtual-screening - Classical docking\n- chemoinformatics\u002Fpose-validation - PoseBusters QC\n- chemoinformatics\u002Ffree-energy-calculations - Post-docking FEP\n- chemoinformatics\u002Fcovalent-design - Covalent docking\n- chemoinformatics\u002Fprotac-degraders - Ternary complex prediction\n\n## Cross-references\n\nOther skills in this category:\n\n- admet-prediction\n- conformer-generation\n- covalent-design\n- free-energy-calculations\n- generative-design\n- molecular-descriptors\n- molecular-io\n- molecular-standardization\n- pharmacophore-modeling\n- pose-validation\n- protac-degraders\n- qsar-modeling\n- retrosynthesis\n- scaffold-analysis\n- shape-similarity\n- similarity-searching\n- substructure-search\n- virtual-screening\n\n## Changelog\n\n- **1.1.0** (migration) — Bulk-migrated to v0.4.0 schema: canonical\n  metadata block, base Hard rules, Cross-references. Body content\n  unchanged; author should review and fill in any domain-specific\n  extensions to the Hard rules.\n- **1.0.0** — Initial release.\n\n",{"name":5,"license":15,"description":7},"main","9c0b41e2036cc5588c699f06e658ea6dc3b9685d","skills\u002Fchemoinformatics\u002Fdocking-rescoring",2,"https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar","PhD Scholar,\r\nManipal School Of Life Sciences, Manipal\r\n #OpenSourceDeveloper #LifeSciences #ComputationalBiology","2026-10-30 01:30:52",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add pradyumnasagar\u002Fopen-research-skills --skill docking-rescoring",{"tool":38,"command":39},"bunx","bunx skills add pradyumnasagar\u002Fopen-research-skills --skill docking-rescoring",{"tool":41,"command":42},"pnpm","pnpm dlx skills add pradyumnasagar\u002Fopen-research-skills --skill docking-rescoring","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Fskills\u002Fchemoinformatics\u002Fdocking-rescoring\u002FSKILL.md",true,[],[47,54,60,67,74,81],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M75KTS5K9QNX8FQBTJW2GQSH","good-skill","A well-formed test fixture for the validation script.","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fgood-skill\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M75KTR7FAT0MMMED2EGC9WZZ","BadSkill","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fbad-skill\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M75KTN949EDF77GJDS8H6VR8","my-new-skill","One sentence (max 200 chars) — what the skill does and when to use it.","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Ftemplate\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M75KTHTFS121SSF1Q5X2FXDA","thesis-structure","Use when planning or checking the architecture of a PhD \u002F MPhil thesis in the life sciences or bioinformatics — monograph vs article-based, chapter sequence, IMRaD-based chapters, abstract-to-conclusion flow, and the examiner expectations that the structure must meet.","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Fskills\u002Fthesis-writing\u002Fthesis-structure\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M75KTE3WKW46FXFHQ579H8YP","thesis-methods-reproducibility","Use when writing or checking the Methods, figures, data, and code of a bioinformatics \u002F microbiology \u002F life-science PhD thesis — the FAIR data, versioned code, report-generation, figure-and-table discipline, and the reproducibility expectations that examiners probe at viva.","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Fskills\u002Fthesis-writing\u002Fthesis-methods-reproducibility\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M75KTA033ZQCJY3DC8P5VQET","thesis-defense-viva","Use when preparing for the viva voce \u002F oral thesis defense — building the examiner's question set, rehearsing the presentation, and the pre-defense polish (slides, summary, the 'what if I'm stuck' responses) for a life-science or bioinformatics PhD.","https:\u002F\u002Fgithub.com\u002Fpradyumnasagar\u002Fopen-research-skills\u002Fblob\u002Fmain\u002Fskills\u002Fthesis-writing\u002Fthesis-defense-viva\u002FSKILL.md",[],[]]