boltz-1

Boltz-1

Boltz-1 is a diffusion-based deep learning model for predicting 3D structures of biomolecular complexes. It can be run on both CPU and GPU systems.

Basic Usage

Boltz-1 can be run with minimal parameters as shown in the examples below.

Create dataset

mkdir -p test_data

# Create a minimal MSA file
cat > test_data/test.a3m << EOF
>101
MVKG
>UniRef100_test
MVKG
EOF

# Create a simple test FASTA with MSA reference
cat > test_data/simple_test.fasta << EOF
>A|protein|./test_data/test.a3m
MVKG
EOF

Test CPU

# Test with minimal parameters and CPU
boltz predict test_data/simple_test.fasta \
        --accelerator cpu \
        --sampling_steps 25 \
        --recycling_steps 1 \
        --diffusion_samples 1 \
        --cache ./boltz_cache

Test GPU

Note: This test requires a CUDA-capable GPU. If no GPU is available, use the CPU version instead.

# Test with minimal parameters and GPU
boltz predict test_data/simple_test.fasta \
        --accelerator gpu \
        --sampling_steps 25 \
        --recycling_steps 1 \
        --diffusion_samples 1 \
        --cache ./boltz_cache

Using Your Own Data

Example with existing data (change path/to/your_protein.fasta to your data):

# Run prediction on your existing FASTA file
boltz predict path/to/your_protein.fasta \
        --accelerator gpu \
        --sampling_steps 25 \
        --recycling_steps 1 \
        --diffusion_samples 1 \
        --cache ./boltz_cache