BindcraftAA: Gradient-Based Protein Binder Design with AlphaFold3

⚠️ Status: Still under active development — use with caution.

---

Overview

BindcraftAA enables protein binder design using gradient-based optimization through AlphaFold3's Evoformer and distogram heads. It implements a gradient-driven binder design pipeline built on top of AlphaFold3, following a methodology similar to BoltzDesign, adapted within the BindCraft framework. The goal is to iteratively optimize a protein sequence to form a high-quality binding interface with a specified target.

---

Features

• Gradient-based sequence optimization
• Integration with AlphaFold3 internal representations
• Supports protein targets (via PDB input) and small-molecule ligands (via CCD codes)

---

Installation

1. Install AlphaFold3 following the official instructions. Make sure you have access to the AF3 weights, databases, and required dependencies.

   • Download AF3 model weights as instructed in the official AlphaFold3 repository.
   • Place your weights file (e.g. af3.bin.zst) in the alphafold3/models directory.
   • See the configs and argument list in our pipeline for full details on required paths.

2. Create the conda environment:

   chmod +x setup.sh
   ./setup.sh

---

Usage

Design a binder against a protein target

python3 bindcraft_af3_gradient.py \
    --target_pdb my_target.pdb \
    --target_chain A \
    --binder_length 100 \
    --num_soft_iterations 80

Design a binder against a small-molecule ligand

python3 bindcraft_af3_gradient.py \
    --ligand_ccd SAM \
    --binder_length 150

---

Citation

If you use this in your research, please cite:

• AlphaFold3 — Abramson et al., 2024
• BoltzDesign — Yehlin et al., 2025
• Original BindCraft — Pacesa et al., 2025