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

Abstract

We present MicroVerse, a software application based on large language models (LLMs) designed to generate C# scripts simulating three-dimensional colonies of model organisms, including Escherichia coli and Saccharomyces cerevisiae, within the Unity platform. The system, which utilizes fine-tuned LLMs, translates descriptions of desired simulations (organizations, growth rates, and appearance) in natural language into functional C# code. The coupling of artificial intelligence with biological simulation generates intricate dynamics to be simulated. The approach involves using a dual-LLM pipeline (one computing user requests, the other generating code for Unity Data-Oriented Technology Stack (DOTS)), with the goal of making 3D simulation development more democratized. It provides a scalable, customizable option, particularly for users lacking sophisticated programming skills, decreasing manual coding effort and time. This work explains the pipeline architecture and fine-tuning process and demonstrates the system successfully producing functional, scalable simulations, scaling up to cases with more than one million organisms.