Skip to content

Quick Start

Get up and running with synth-pdb in 5 minutes.

Installation

pip install synth-pdb

git clone https://github.com/elkins/synth-pdb.git
cd synth-pdb
pip install .

Your First Structure

Generate a simple 10-residue alpha helix:

synth-pdb --length 10 --conformation alpha --output my_first_helix.pdb

This creates a PDB file with: - ✅ Full atomic representation (backbone + side-chains + hydrogens) - ✅ Realistic B-factors and occupancy values - ✅ Proper bond geometry and angles

Visualize It

View your structure interactively in the browser:

synth-pdb --length 10 --conformation alpha --visualize

Add Physics

Generate a more realistic structure with energy minimization:

synth-pdb --length 20 --conformation alpha --minimize --output minimized.pdb

This uses OpenMM to: - Regularize bond lengths and angles - Resolve steric clashes - Apply implicit solvent effects

Next Steps

  • First Structure Tutorial

    Detailed walkthrough of generating your first protein structure

    Start tutorial

  • Command-Line Reference

    Complete reference of all available options and flags

    View reference

  • Examples Gallery

    Browse inspiring examples and copy-paste commands

    Explore gallery

  • AI/ML Integration

    Learn how to use synth-pdb for machine learning workflows

    ML guide

Common Use Cases

Testing Bioinformatics Tools

# Generate a test dataset
synth-pdb --mode dataset --num-samples 100 --output ./test_data

Educational Demonstrations

# Show different secondary structures
synth-pdb --structure "1-10:alpha,11-20:beta" --visualize

NMR Data Generation

# Generate structure with synthetic NMR observables
synth-pdb --length 30 --gen-nef --gen-relax --output nmr_test.pdb

Synthetic Data Factories

For high-throughput AI training, synth-pdb includes a dedicated builder for multi-modal datasets:

# Generate 500 synchronized PDB, Cryo-EM, SAXS, and NMR samples
python3 scripts/build_multimodal_dataset.py --n 500 --output-dir my_ai_dataset

This factory leverages the 900x faster vectorized walker to produce labeled training data in minutes.

Getting Help

Need help?