Prompt-Only Gene-Circuit Modeling with a Large Language Model Simulates the synNotch Spheroids

Area of Science:

• Computational Biology
• Developmental Biology
• Systems Biology

Background:

• Quantitative simulations of morphogenesis are crucial for understanding gene regulatory circuits.
• Constructing these simulations is technically demanding, posing a barrier for many researchers.

Purpose of the Study:

• To develop a prompt-only pipeline using large language models (LLMs) to automatically generate CompuCell3D (CC3D) simulations from natural-language biological specifications.
• To assess the LLM's ability to capture key biological dynamics in simulations without manual programming.

Main Methods:

• Utilized a large language model (LLM) to generate CompuCell3D (CC3D) code from natural language descriptions.
• Employed the synNotch spheroid assay as a benchmark for validating simulation accuracy.
• Iteratively refined simulation parameters, such as incorporating a cumulative contact-duration threshold, to address observed discrepancies.

Main Results:

• LLM-generated CC3D code successfully simulated contact-dependent signaling, cadherin-mediated adhesion, and multicellular sorting dynamics.
• Initial simulations showed spheroid fragmentation, indicating premature signaling as a failure mode.
• Implementing a contact-duration threshold rescued single-spheroid formation, highlighting the importance of sustained cell-cell interactions.

Conclusions:

• Conversational LLM guidance significantly accelerates multiscale modeling by simplifying the CompuCell3D (CC3D) workflow.
• This framework empowers nonspecialists to test gene-circuit hypotheses in silico.
• The approach offers a generalizable method for rapid, mechanism-level exploration of developmental design principles.