MP4

The Molecule Programming Foundational Model (MP4) from 310 AI is a generative AI system designed to create high-quality, de novo protein sequences from text prompts. By compressing 4 billion years of natural evolution into a single powerful model, MP4 is built to transform how researchers, engineers, and life scientists study and utilize biology — opening new possibilities across medicine, industrial biotechnology, environmental engineering, and fundamental research.

MP4 is intended for scientists and organizations seeking to accelerate protein design beyond the limitations of traditional, non-AI methods. The platform provides access to growing repositories of AI-generated proteins and enzymes, each rigorously evaluated for quality and applicability, making it a practical tool for both discovery and applied molecular programming.

Core Capabilities

• Generates high-quality de novo proteins from natural language text prompts, enabling intuitive and flexible molecular design workflows.
• Supports applications across multiple domains, including medicine, industrial processes, and environmental engineering.
• Produces diverse protein types, including enzymes, scaffolds, and hormones.
• Each AI-generated protein is evaluated for amino acid composition, structural confidence, and functional similarity to ensure effectiveness and real-world applicability.

Protein and Enzyme Repositories

• The MP4 introductory repository contains over 1,053 AI-designed proteins, with the collection continuing to grow over time.
• The MP Enzyme Commission (EC) repository includes 6,590 AI-generated proteins, achieving 100% coverage of top-level enzyme classes in EC space.
• The EC repository includes 725 never-before-seen enzyme discoveries, representing genuinely novel contributions to the known protein landscape.

Validated Lab Performance

• MP4 achieved an 84% lab expression rate, vastly surpassing the 20–30% expression rates typically seen with non-AI protein design methods.
• In a demonstrated case study, MP4 designed a 214-amino-acid enzyme with potential applications for heart disease treatment.
• The AI-designed enzyme exhibits substantial sequence divergence from natural proteins, with 52 mutations from its closest natural sequence match, while maintaining a very high structural similarity score of 0.98 out of 1.00.
• The designed protein notably omits a conserved zinc-binding loop found in natural counterparts, illustrating the model's ability to explore novel structural solutions beyond evolutionary constraints.

310 AI provides white papers and dedicated repositories for each major MP4 capability area, allowing researchers to explore the underlying methodology and access generated protein datasets directly. The platform represents a significant step forward in programmable protein design, combining generative AI with rigorous experimental validation.