VLGPO: Variational Latent Generative Protein Optimization

ICML 2025

This repository contains the inference code for A Variational Perspective on Generative Protein Fitness Optimization accepted at ICML 2025.

Lea Bogensperger¹, Dominik Narnhofer², Ahmed Allam¹, Konrad Schindler², Michael Krauthammer¹

¹University of Zurich, ²ETH Zurich

---

🛠️ Setup

Checkpoints for the predictors $g_\phi$ and $g_{\tilde{\phi}}$ (classifier guidance) and the in-silico oracle $g_\psi$ (evaluation) are taken from GGS: Gibbs sampling with Graph-based Smoothing.

📦 Repository

The file requirements.txt contains a list of the Python packages required to run this project.

🚀 Run code

# 1. Create and activate the environment
conda create -n vlgpo-env python=3.11
conda activate vlgpo-env
pip install -r requirements.txt

# 2. Run the sampler
python src/vlgpo/sample.py

🎓 Citation

@article{bogensperger2025variational,
  title={A Variational Perspective on Generative Protein Fitness Optimization},
  author={Bogensperger, Lea and Narnhofer, Dominik and Allam, Ahmed and Schindler, Konrad and Krauthammer, Michael},
  journal={arXiv preprint arXiv:2501.19200},
  year={2025}
}