An LLM-Based Tool for Automated Generation of 3D Model Organism Colony Simulations

Area of Science:

• Computational Biology
• Bioinformatics
• Software Engineering

Background:

• Developing complex 3D biological simulations, particularly for microorganisms like *Escherichia coli* and *Saccharomyces cerevisiae*, requires significant programming expertise.
• Existing simulation tools often have steep learning curves, limiting accessibility for researchers without specialized coding skills.

Purpose of the Study:

• To introduce MicroVerse, an AI-powered software application that simplifies the creation of 3D biological simulations.
• To democratize the development of intricate, large-scale simulations by leveraging natural language processing and artificial intelligence.

Main Methods:

• Utilized fine-tuned large language models (LLMs) within a dual-LLM pipeline architecture.
• One LLM interprets user requests in natural language, while the second generates C# scripts for the Unity Data-Oriented Technology Stack (DOTS).
• Developed a system capable of translating simulation parameters (organization, growth, appearance) into functional code.

Main Results:

• Successfully generated functional and scalable C# scripts for simulating 3D colonies of model organisms.
• Demonstrated the system's ability to create simulations with over one million individual organisms.
• Validated the effectiveness of the AI-driven approach in reducing manual coding effort and development time.

Conclusions:

• MicroVerse successfully democratizes 3D biological simulation development through an AI-driven pipeline.
• The application offers a scalable and customizable solution, empowering users with limited programming experience.
• This approach integrates artificial intelligence with biological simulation to generate complex, large-scale dynamics efficiently.