[Open Source] HanjariFold2: CosmicDistiller Engine Released! A Memory-Efficient AI Protein Structure Refinement Tool

Hello, Structural Biology & AI Community!

Amidst the intense modeling rounds of CASP17 this week, I am thrilled to officially release HanjariFold2: CosmicDistiller Engine as an open-source software!

CosmicDistiller is an ultra-fast, lightweight macromolecular refinement tool designed to clean up unphysical atom clashes and backbone geometry errors (Ramachandran outliers) commonly generated by state-of-the-art AI models like AlphaFold3, while simultaneously building an optimized hydration layer (explicit water molecules).

GitHub Repository: https://github.com/bghan2024/HanjariFold-endgame
License: Apache License 2.0 (Free for academic and commercial use, citation required)

💡 Why CosmicDistiller? (Background)

While cutting-edge AI models have revolutionized protein structure prediction, they often struggle with exact physical constraints—frequently spitting out unphysical atom clashes and distorted backbone geometries (Ramachandran outliers). Furthermore, these AI models typically predict structures in a “dry” state, lacking essential hydration waters.

Traditional refinement tools require heavy molecular dynamics (MD) environments and easily suffer from Out of Memory (OOM) crashes when dealing with large complexes.

CosmicDistiller elegantly solves this bottleneck with an exceedingly clean, ~100-line Python architecture without compromising physical rigor.

✨ Key Features

Macromolecular Scalability (No More OOM!) By incorporating a highly optimized 3D grid bounding-box algorithm with a 4.5Å grid spacing, it easily refines massive macromolecular complexes like Ribosomes or Viral Capsids within seconds on standard lab workstations—completely bypassing memory crashes.
Drastic Quality Improvement (Data-Proven) It rigorously purges geometric distortions. Real-world target testing demonstrates that it wipes out Ramachandran outliers down to 0.00% and condenses the structure to a top-tier MolProbity Score in the 1.0 range (comparable to ~1.0Å ultra-high-resolution experimental structures).
Ultra-High Compatibility & Interoperability It fluidly ingests both legacy PDB files and next-generation mmCIF formats, exporting flawless, production-ready mmCIF data models.

🚀 Easy 3-Step Guide (How to Use)

As long as you have Phenix installed—a standard in almost every structural biology lab—you are ready to go. No dependency hell, no version conflicts.

Step 1: Initialize Phenix Environment

Open your terminal window and source your Phenix environment variables:

Bash

source /home/hanjari/phenix-2.0-5936/phenix_env.sh
# (*Please modify the path according to your local Phenix installation!)

Step 2: Run the Engine

Navigate to the directory containing the script, type phenix.python followed by the script name and your target structure file path, then hit Enter:

Bash

phenix.python HanjariFold2_CosmicDistiller.py your_protein.cif

Step 3: Verify the Results

The engine will instantly generate a folder named HF2_DISTILLER_YYYYMMDD_HHMMSS. Inside, you will find your beautifully distilled, physically error-free structure file (distilled_final.cif).

🚨 License & Mandatory Citation Rules

This project is distributed under the Apache License 2.0, allowing total freedom for modification, academic research, and commercial exploitation. Notably, for commercial deployment, there is absolutely no obligation to open-source your proprietary code (No Code-Open Duty), provided you fully comply with the citation guidelines below.

⚠️ [Mandatory Condition] Strict Citation Policy

Whether you are an individual researcher, an academic institute, or a commercial biopharma enterprise, if you integrate CosmicDistiller into your pipelines, publish papers, file patents, or commercialize derived therapeutic candidates, you MUST explicitly cite the following official source:

Markdown

Hanjari, et al. (2026). "Memory-Efficient Continuum Hydration for Ultra-Large Macromolecular Refinement." Journal of Molecular Biology (or CASP17 Proceedings).

⚠️ [Crucial Notice] Third-Party Dependency (Phenix) License Clause

While the CosmicDistiller script itself is licensed under the Apache License 2.0, the license for the underlying Phenix software suite invoked by this tool remains entirely separate.

Please ensure you adhere to the Phenix Consortium license regulations:

Academic/Non-profit Users: May request and utilize Phenix licenses free of charge.
Commercial/For-profit Users (Biopharma, Startups, etc.): You MUST independently acquire a valid Phenix Commercial License to run this pipeline for commercial purposes. The author of this repository accepts no liability for any legal non-compliance regarding third-party software licensing.

🌐 Official Phenix Licensing & Download: https://phenix-online.org/download/

👩‍💻 Developer Info

Author: Han Byeong-gu
Email: hanbyeonggu@gmail.com
GitHub Repository: bghan2024/HanjariFold-endgame

Visit our GitHub repository today, check out the code, and don’t forget to drop a Star (⭐) if you find it useful! Let’s push the boundaries of structural biology together. Thank you!