De Novo Binder Design

Design protein binders using BindCraft with GPU-accelerated deep learning.

Table of contents

1. Overview
2. Required MCPs
3. Configuration
   1. Target PDB Requirements
4. Pipeline Steps
   1. Step 1: Explore Configurations
   2. Step 2: Setup Results Directory
   3. Step 3: Generate Configuration
   4. Step 4: Submit Design Job
   5. Step 5: Monitor Progress
   6. Step 6: Get Results
   7. Step 7: Visualize Results
5. Output Structure
6. Quality Thresholds
7. Run the Workflow

---

Overview

This workflow designs de novo protein binders against a target protein using BindCraft’s integrated pipeline:

1. AF2 Hallucination — Generate binder backbone conformations guided by target structure
2. MPNN Sequence Design — Optimize amino acid sequences for designed backbones
3. AF2 Prediction — Validate binder-target complexes with structure prediction
4. PyRosetta Analysis — Score interface quality, energy, and structural metrics

Required MCPs

pskill install binder_design

This installs: bindcraft_mcp

Configuration

TARGET_PDB: "examples/data/target.pdb"     # Target protein PDB file
TARGET_CHAINS: "A"                          # Target chains to design binders for
BINDER_LENGTH: 130                          # Length of designed binder
RESULTS_DIR: "results/binder_design"        # Output directory
NUM_DESIGNS: 3                              # Number of designs to generate
HOTSPOT_RESIDUES: null                      # Optional: specific residues to target

Target PDB Requirements

• Clean PDB file with target protein structure
• All heteroatoms and waters removed (unless needed)
• Chain IDs properly assigned

Pipeline Steps

Step 1: Explore Configurations

Use generate_config to see available settings and analyze the target structure.

Step 2: Setup Results Directory

Create the output directory structure.

Step 3: Generate Configuration

Analyze the target PDB and generate an optimized BindCraft configuration including target_settings.json, design filters, and advanced settings.

Step 4: Submit Design Job

Submit an asynchronous design job. BindCraft runs RFdiffusion for backbone generation, ProteinMPNN for sequence design, and AlphaFold2 for validation.

Step 5: Monitor Progress

Check job status using bindcraft_check_status. The job reports completed trajectories, accepted designs, and rejected designs.

Step 6: Get Results

Retrieve completed designs with quality metrics (pLDDT, pAE, interface scores).

Step 7: Visualize Results

Generate publication-ready figures:

• pLDDT comparison bar chart
• Interface pAE scores
• Quality scatter plot (pLDDT vs pAE)
• Design ranking by composite score
• Metrics summary table
• Execution timeline

Output Structure

RESULTS_DIR/
├── config/
│   ├── target_settings.json
│   ├── default_filters.json
│   └── job_output/
│       ├── Accepted/Ranked/       # Final ranked PDB designs
│       ├── Rejected/
│       ├── final_design_stats.csv
│       └── bindcraft_run.log
└── designs/

Quality Thresholds

Metric	Threshold	Direction
pLDDT	≥ 80	Higher is better
pAE	≤ 5	Lower is better
i_pAE (interface)	≤ 10	Lower is better
i_pTM (interface)	≥ 0.6	Higher is better

Run the Workflow

pskill install binder_design
claude
> /binder-design