ALGORITHM_INDEX.md

Boltz-2 Algorithm Index

This index covers the new and enhanced algorithms in Boltz-2 compared to Boltz-1.

Key Innovations in Boltz-2

Feature	Description
Binding Affinity Prediction	First DL model approaching FEP accuracy, 1000x faster
Dual Output	affinity_pred_value (log10 IC50) + affinity_probability_binary
Contact Conditioning	Guide predictions with experimental contacts
Template v2	Enhanced template processing
Improved Confidence	Better uncertainty estimation

Architecture Comparison

┌─────────────────────────────────────────────────────────────────────┐
│                    Boltz-2 vs Boltz-1                               │
├─────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  Boltz-1:  Input → MSA → Pairformer → Diffusion → Structure         │
│                                                    ↓                 │
│                                               Confidence             │
│                                                                      │
│  Boltz-2:  Input → MSA → Pairformer → Diffusion → Structure         │
│              ↓                                     ↓                 │
│         Templates v2                          Confidence v2          │
│         Contact Cond.                              ↓                 │
│                                             ┌──────────────┐         │
│                                             │ Affinity     │ ← NEW!  │
│                                             │ Module       │         │
│                                             └──────────────┘         │
│                                                                      │
└─────────────────────────────────────────────────────────────────────┘

Algorithms

New in Boltz-2

#	Algorithm	Notebook	Source File	Status
1	Affinity Module	algorithm-01-AffinityModule.ipynb	model/modules/affinity.py	✅
2	Gaussian Smearing	algorithm-02-GaussianSmearing.ipynb	model/modules/affinity.py	✅
3	Contact Conditioning	algorithm-03-ContactConditioning.ipynb	model/modules/trunkv2.py	✅
4	Affinity Heads Transformer	algorithm-04-AffinityHeadsTransformer.ipynb	model/modules/affinity.py	✅

Enhanced in Boltz-2 (v2 modules)

#	Algorithm	Notebook	Source File	Status
5	Input Embedder v2	algorithm-05-InputEmbedderV2.ipynb	model/modules/trunkv2.py	✅
6	Template Module v2	algorithm-06-TemplateModuleV2.ipynb	model/modules/trunkv2.py	✅
7	Diffusion v2	algorithm-07-DiffusionV2.ipynb	model/modules/diffusionv2.py	✅
8	Confidence v2	algorithm-08-ConfidenceV2.ipynb	model/modules/confidencev2.py	✅
9	Distogram v2	algorithm-09-DistogramV2.ipynb	model/loss/distogramv2.py	✅
10	B-Factor Prediction	algorithm-10-BFactorPrediction.ipynb	model/modules/trunkv2.py	✅

Source Code References

• Official Repository: jwohlwend/boltz (contains both Boltz-1 and Boltz-2)
• Boltzina (Virtual Screening): ohuelab/boltzina
• Boltz-2 Paper: bioRxiv 2025.06.14.659707

Usage

Boltz-2 is the default model when running:

boltz predict input.yaml --use_msa_server

For affinity prediction, use a YAML with affinity specifications:

sequences:
  - protein:
      id: A
      sequence: MVLSPADKTN...
  - ligand:
      id: B  
      smiles: CC(=O)NC1=CC=C(O)C=C1
affinity:
  predict: true