MutexaGPT: An Intuition-to-Design Translator for Physics-based Enzyme Engineering

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Protein Engineering

Background:

• Enzyme engineering relies on physical intuitions but lacks systematic methods to translate these into quantitative design principles.
• Existing approaches struggle to bridge qualitative insights with actionable, physics-based enzyme designs.

Purpose of the Study:

• To introduce MutexaGPT, an open-access, multi-agent large language model (LLM) platform.
• To enable the translation of enzyme engineering intuition into physics-based simulations and variant designs.
• To democratize physics-guided, intuition-driven enzyme engineering.

Main Methods:

• MutexaGPT utilizes a web interface to accept plain English intuition-driven requests.
• LLM agents (QuestionAnalyzer, WorkPlanningBoard, ResultExplainer) process requests, build physics-based models, and execute molecular modeling workflows.
• An automated evaluation framework benchmarks prompt engineering strategies for agent optimization.

Main Results:

• MutexaGPT successfully engineered halide methyltransferase for bulkier substrates, achieving a 40% hit rate and 4-fold activity improvement.
• MutexaGPT generated cold-adapted amylase variants with 1.7-fold and 3.7-fold activity enhancement at 0 °C.
• The platform demonstrated effective translation of intuition into actionable design proposals, such as smart mutation libraries.

Conclusions:

• MutexaGPT serves as an effective intuition-to-design translator for enzyme engineering.
• The platform integrates human creativity with high-throughput molecular modeling.
• MutexaGPT democratizes the process of physics-guided, intuition-driven enzyme design.