docking-rescoring

Version Compatibility

Reference 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+.

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures
• CLI: diffdock --version; boltz --version

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

ML Docking and Rescoring

Use machine learning models for protein-ligand pose prediction and affinity scoring. The field underwent a major shift in 2023-2025: foundation models (AlphaFold3, Boltz-1, Chai-1) handle protein-ligand prediction natively; diffusion-based docking (DiffDock-L) generates poses; Boltz-2 affinity module approaches FEP accuracy at 1000x speed. Critical caveat: PoseBusters (Buttenschoen 2024) showed ML methods produce ~50% physically-invalid poses despite RMSD <= 2 Å; classical methods (Vina, GOLD) produce ~5-15% invalid. The postdoc-grade workflow is hybrid: ML for pose sampling + classical rescoring + physical validation.

For classical docking, see chemoinformatics/virtual-screening. For pose validation (PoseBusters), see chemoinformatics/pose-validation. For free-energy calculations (post-docking), see chemoinformatics/free-energy-calculations. For PROTAC ternary complex prediction, see chemoinformatics/protac-degraders.

ML Docking Method Taxonomy

Tool	Approach	Speed	Strength	Fails when
DiffDock-L (Corso 2024)	Equivariant diffusion	5s/lig GPU	Pose sampling for cross-dock	~50% PB-invalid; OOD
Boltz-1 (Wohlwend 2024)	AlphaFold-style foundation	10s GPU	Full complex prediction	DNA / RNA may be off
Boltz-2 (Wohlwend 2025)	Boltz-1 + affinity head	10s GPU	Pose + affinity (Pearson 0.66 on 4-target FEP+ subset; RMSE ~1.5 kcal/mol on ChEMBL holdout)	Novel chemotype OOD
Chai-1 (Chai 2024)	AlphaFold-style + LM	10s GPU	Pose 77% RMSD success on PoseBusters	Limited public
AlphaFold3 (DeepMind 2024)	Foundation model	API only	Pose 76% RMSD on PoseBusters	Restricted API access
EquiBind	Equivariant single-shot	<1s GPU	Fast pose	Lowest accuracy on PoseBusters
TANKBind	Distance + classifier	<1s GPU	Fast pose + score	Geometric inconsistency
NeuralPLexer	E3-equivariant	<1s	Fast pose	Limited adoption
Glide (Schrödinger)	Hybrid grid + ML rescoring	30s GPU	Commercial SOTA	License cost
GNINA 1.1 CNN	Classical sampling + CNN scoring	30s GPU	Best classical-hybrid	Limited to PDBbind chemotypes

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.

Decision Tree by Scenario

Scenario	Recommended workflow
Known holo, need fast pose	GNINA classical
Apo or AF-predicted protein, need pose	Boltz-1 or Chai-1
Cross-docking + scaffold hopping	DiffDock-L + GNINA rescore + PoseBusters
Affinity prediction (replace FEP first-pass)	Boltz-2 affinity module
Ultralarge library (1M+)	Vina pre-filter -> GNINA on top 1% -> Boltz-2 on top 0.1%
Novel target family	Boltz-1 / Chai-1 (uses MSA flexibility)
Cofactor / metal binding	AlphaFold3 (best cofactor handling); validate with classical
PROTAC / bivalent	Boltz-1 / Chai-1 with multimer + constraints
Production with auditable poses	GNINA classical + Boltz-2 score

PoseBusters Problem (Critical)

PoseBusters benchmark (Buttenschoen 2024) showed:

Tool	RMSD <= 2 Å	PB-valid	RMSD <= 2 Å AND PB-valid
Vina (default)	65%	90%	60%
GOLD	70%	88%	65%
GNINA CNN	73%	85%	65%
DiffDock-L	55%	40%	25%
EquiBind	30%	25%	10%
TANKBind	45%	35%	20%
AlphaFold3 ligand	76%	65%	55%
Chai-1	77%	70%	58%
Boltz-1	74%	68%	55%
Boltz-2 (with affinity)	76%	70%	58%

Conclusion: Modern foundation models match classical RMSD but with worse physical plausibility. Always require PB-valid + RMSD <= 2 Å.

DiffDock-L + GNINA Hybrid Workflow (Production Standard)

Goal: Use DiffDock-L for fast diverse pose sampling; GNINA CNN to rescore; PoseBusters to filter.

# Step 1: DiffDock-L pose sampling
python -m inference \
    --protein_path receptor.pdb \
    --ligand_description smiles.smi \
    --out_dir diffdock_out/ \
    --samples_per_complex 40 \
    --inference_steps 20

# Step 2: GNINA CNN rescoring
gnina -r receptor.pdb -l diffdock_out/top_poses.sdf \
      --autobox_ligand reference_ligand.sdf \
      --cnn_scoring rescore \
      -o gnina_rescored.sdf.gz

# Step 3: PoseBusters QC"
python -c "
from posebusters import PoseBusters
b = PoseBusters(config='dock')
df = b.bust(mol_pred='gnina_rescored.sdf.gz', mol_cond='receptor.pdb')
print(df.filter(regex='pass').sum().sum(), 'checks passed')
"

Boltz-2 Affinity Prediction (Fast FEP Replacement)

Goal: Predict binding affinity at ~10s/ligand on GPU vs hours for FEP.

Approach: Submit protein + ligand to Boltz-2 with the affinity head; receive predicted pIC50 or ΔG with uncertainty.

boltz predict input.yaml --use_affinity --output_dir boltz2_out/

Validation: Boltz-2 reports Pearson 0.66 vs experimental on 4-target FEP+ subset; RMSE ~1.5 kcal/mol on ChEMBL holdout. Faster than FEP (10s vs hours) but ~3x less accurate.

Foundation Models: Boltz-1 and Chai-1

For novel target or no MSA, the foundation models treat ligand + protein jointly:

# Boltz-1
boltz predict input.yaml --output_dir boltz1_out/

# Chai-1
chai-lab predict input.fasta --output-dir chai_out/

Boltz-1 / 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.

Per-Tool Failure Modes

DiffDock-L -- PB-invalid poses

Trigger: Default DiffDock-L output.

Mechanism: Diffusion lacks explicit physical-plausibility term; ~50% of poses fail planarity, vdW overlap, or chirality tests.

Symptom: Poses look reasonable in 2D depiction but fail QC.

Fix: Always apply PoseBusters filter; rescore with GNINA.

Boltz-2 -- poor affinity for novel chemotype

Trigger: Ligand chemotype outside ChEMBL training distribution.

Mechanism: Affinity head trained on public bioactivity data; extrapolation is unreliable.

Symptom: Predicted pIC50 far from experimental; uncertainty band very wide.

Fix: Use Boltz-2 for first-pass triage only; FEP for top hits.

EquiBind -- worst accuracy

Trigger: Default EquiBind without ensemble.

Mechanism: Single-shot equivariant network; no refinement step.

Symptom: Poses within RMSD > 5 Å.

Fix: Avoid EquiBind; use DiffDock-L or Boltz-1 for sampling, then refine with GNINA.

AlphaFold3 -- restricted API

Trigger: Academic users without commercial agreement.

Mechanism: AF3 server requires Google Cloud account + quota.

Symptom: Cannot submit jobs; rate-limited.

Fix: Use Boltz-1 (open source) for similar performance; AF3 only for cross-validation.

ML methods ignore cryptic pockets

Trigger: Binding site not in receptor conformation.

Mechanism: ML methods score poses against static receptor.

Symptom: All poses report poor affinity; known active missed.

Fix: Pre-generate receptor ensemble via MD or use AlphaFold3 holo prediction.

Reconciliation: ML vs Classical

Aspect	Classical (Vina/GNINA)	ML (DiffDock/Boltz)
Pose RMSD	60-70% within 2 Å	50-77% within 2 Å
PB-validity	85-90%	40-70%
Affinity accuracy	Correlates ~0.5 with exp	Boltz-2 ~0.66 Pearson on benchmark
Speed	5-30s/lig	5-10s GPU
Out-of-distribution	Robust for chemotypes	Worse for novel scaffolds
Interpretability	Force-field based	Black-box

Production hybrid: ML for pose sampling (broader search), classical for affinity + physical validation.

Common Errors

Symptom	Cause	Fix
Boltz-2 OOM on big protein	MSA + ligand fits in 16GB	Reduce MSA depth; use Boltz-1
DiffDock all poses cluster	Insufficient samples	Increase samples_per_complex to 40-100
Boltz output not a single chain	Tokenizer confused by modified residues	Strip non-standard residues; use UniProt canonical
GNINA can't read DiffDock output	SDF missing 3D	DiffDock writes 3D SDF; verify with PyMOL
Chai-1 ignores pocket	Wrong binding-site hint	Pass --pocket-residues if available

References

• Corso et al., Nat. Mach. Intell. -- DiffDock-L.
• Wohlwend et al., 2024 -- Boltz-1; 2025 -- Boltz-2 with affinity.
• Buttenschoen et al., Chem. Sci. 15:3130 -- PoseBusters benchmark.
• McNutt et al., J. Cheminformatics 13:43 -- GNINA 1.0 CNN.
• Krishna et al., 2024 -- AlphaFold3 server.

Related Skills

• chemoinformatics/virtual-screening - Classical docking
• chemoinformatics/pose-validation - PoseBusters QC
• chemoinformatics/free-energy-calculations - Post-docking FEP
• chemoinformatics/covalent-design - Covalent docking
• chemoinformatics/protac-degraders - Ternary complex prediction