Workflow · Production

Boltz: Open-Source Generative Protein Structure Prediction and Design with Boltz-1, Boltz-2, and Boltz Lab

The problem

While single-chain protein structure prediction had seen dramatic progress, modeling complex molecular interactions—protein-ligand, protein-protein—and enabling generative protein design remained open challenges critical to drug discovery and biology.

First attempt

Prior regression-based structure prediction models produced averaged outputs when the ground truth was ambiguous between multiple valid conformational states, rather than sampling the full posterior distribution of possible structures.

Workflow diagram · grounded in source
1
User submits design spec
trigger
“users feed the model blank tokens and a high-level "spec" (e.g., an antibody framework), and the model decodes both the 3D structure and the corresponding amino acids”
2
Unified structure-sequence encoding
ai_action
“Boltz-2 (and BoltzGen) treats structure and sequence prediction as a single task by encoding amino acid identities into the atomic composition of the predicted structure”
3
Generative structure and sequence decoding
ai_action
“the model decodes both the 3D structure and the corresponding amino acids”
4
Affinity prediction
ai_action
“Boltz-2 focuses on affinity prediction—quantifying exactly how tightly a designed binder will stick to its target”
5
Validation on novel targets
validation
“Boltz tested its designs on 9 targets with zero known interactions in the PDB, achieving nanomolar binders for two-thirds of them”
6
Medicinal chemist review
human_review
“allowing them to run parallel screens and use their intuition to filter model outputs”
Reported outcome

Boltz released open-source Boltz-1 and Boltz-2 models approaching AlphaFold 3 performance, and launched Boltz Lab—a platform running 10x faster than open-source versions—that achieved nanomolar binders for two-thirds of 9 novel validation targets with zero prior PDB interactions.

Reported metrics
Boltz Lab inference speed vs open-source10x faster
Nanomolar binders achieved on novel validation targetstwo-thirds
novel PDB-zero-interaction targets tested9
Reported stack
Boltz-1Boltz-2BoltzGenBoltz LabProtein Data Bank
Source
https://www.latent.space/p/boltz
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Boltz released open-source Boltz-1 and Boltz-2 models approaching AlphaFold 3 performance, and launched Boltz Lab—a platform running 10x faster than open-source versions—that achieved nanomolar binders for two-thirds…

What tools did this team use?

Boltz-1, Boltz-2, BoltzGen, Boltz Lab, Protein Data Bank.

What results were reported?

Boltz Lab inference speed vs open-source: 10x faster; Nanomolar binders achieved on novel validation targets: two-thirds; novel PDB-zero-interaction targets tested: 9 (source-reported, not independently verified).

What failed first in this deployment?

Prior regression-based structure prediction models produced averaged outputs when the ground truth was ambiguous between multiple valid conformational states, rather than sampling the full posterior distribution of po…

How is this workflow AI workflow structured?

User submits design spec → Unified structure-sequence encoding → Generative structure and sequence decoding → Affinity prediction → Validation on novel targets → Medicinal chemist review.