Can I run Boltz locally?

Yes, the skill supports local installation via pip and requires a CUDA-compatible GPU with at least 24GB of VRAM, though 48GB is recommended for larger complexes.

Is Boltz faster than other predictors?

Yes, Boltz (especially Boltz-2) is optimized for speed and can be significantly faster than AlphaFold2-Multimer, particularly when configured to run without an MSA server.

What is the difference between Boltz and AlphaFold2?

Boltz is an open-source biomolecular predictor that utilizes diffusion models and offers an MSA-free mode, often resulting in faster predictions than AlphaFold2 while maintaining high accuracy for complexes.

Does Boltz support protein-ligand complexes?

Yes, both Boltz-1 and Boltz-2 are designed to predict interactions between proteins and small molecule ligands, making them useful for drug discovery workflows.

What output format does Boltz use?

Boltz provides structure predictions in CIF format along with a JSON file containing confidence metrics like pLDDT, pTM, and ipTM, plus a PAE matrix.

Boltz Structure Prediction

Name: Boltz Structure Prediction
Author: adaptyvbio

byadaptyvbio

•

Ciencia de Datos y ML

Predicts biomolecular structures and protein-ligand complexes using the open-source Boltz-1 and Boltz-2 models.

Acerca de

The Boltz skill integrates the Boltz-1 and Boltz-2 biomolecular structure predictors into Claude Code, enabling researchers to predict protein complexes, validate designed binders, and model protein-ligand interactions. As an open-source alternative to AlphaFold2, it offers faster performance, MSA-free modes, and diffusion-based sampling, making it ideal for large-scale validation campaigns and local GPU execution via Modal or direct installation.

Características Principales

75 GitHub stars
Generates comprehensive confidence metrics including pLDDT, pTM, and ipTM
Scalable execution via Modal cloud GPUs or local CUDA environments
Supports both Boltz-1 and high-performance Boltz-2 models
Offers MSA-free mode for significantly accelerated prediction speeds
Predicts protein-protein and protein-ligand complex structures

Casos de Uso

Predicting the 3D geometry of protein-ligand complexes for drug discovery
High-throughput structural screening of large protein libraries
Validating de novo designed protein binders against target proteins

Acerca de

Características Principales

75 GitHub stars
Generates comprehensive confidence metrics including pLDDT, pTM, and ipTM
Scalable execution via Modal cloud GPUs or local CUDA environments
Supports both Boltz-1 and high-performance Boltz-2 models
Offers MSA-free mode for significantly accelerated prediction speeds
Predicts protein-protein and protein-ligand complex structures

Casos de Uso

Predicting the 3D geometry of protein-ligand complexes for drug discovery
High-throughput structural screening of large protein libraries
Validating de novo designed protein binders against target proteins