Revolutionizing Protein Engineering: The AiCE Approach Integrates AI with Structural and Evolutionary Constraints

A groundbreaking method developed by researchers at the Chinese Academy of Sciences promises to significantly advance the field of protein engineering. This innovative approach, termed AI-informed Constraints for protein Engineering (AiCE), utilizes artificial intelligence to seamlessly integrate structural and evolutionary constraints into a generic inverse folding model, thus obviating the need for specialized AI models. The study, led by Professor Gao Caixia from the Institute of Genetics and Developmental Biology (IGDB), was published on July 7, 2025, in the prestigious journal Cell.

Traditional protein engineering techniques have often faced challenges related to cost, efficiency, and scalability. Conventional AI-based approaches can be computationally intensive, which limits their accessibility to a broader research community. AiCE aims to address these issues by providing a more user-friendly alternative that maintains predictive accuracy and facilitates the evolution of proteins in a rapid and efficient manner.

The AiCE framework comprises two main modules: AiCE single and AiCE multi. The AiCE single module is designed to predict high-fitness single amino acid substitutions. It achieves improved prediction accuracy by extensively sampling inverse folding models—AI models that generate compatible amino acid sequences based on protein 3D structures—while incorporating structural constraints. Notably, benchmarking against 60 deep mutational scanning (DMS) datasets revealed that AiCE single outperforms existing AI methods by an impressive 36% to 90%.

Moreover, the researchers also introduced the AiCE multi module, which integrates evolutionary coupling constraints to predict multiple high-fitness mutations with minimal computational cost. This advancement significantly enhances the practical utility of the AiCE framework. During their research, the team successfully evolved eight proteins with diverse structures and functions, including deaminases and nuclear localization sequences. These engineered proteins have led to the development of next-generation base editors, which are poised to have significant implications in precision medicine and molecular breeding.

Among the notable products of this research are enABE8e, an advanced cytosine base editor with a narrower editing window; enSdd6-CBE, an adenine base editor that exhibits 1.3 times higher fidelity; and enDdd1-DdCBE, a mitochondrial base editor demonstrating a remarkable 13 times increase in activity. The AiCE framework represents a paradigm shift in protein engineering, unlocking the potential of existing AI models to enhance the interpretability and efficiency of AI-driven protein redesign.

This research not only provides insights into the evolution of proteins but also sets the groundwork for future advancements in biotechnology and therapeutic applications. The implications of AiCE extend beyond academia, as the tool's broad applicability could revolutionize industries that rely on protein engineering, including pharmaceuticals, agriculture, and bioengineering. As the field continues to evolve, the integration of AI in protein engineering is likely to become increasingly crucial for the development of novel therapeutics and biotechnological innovations.

In summary, the AiCE framework exemplifies a simple yet powerful strategy for protein engineering that combines structural and evolutionary insights with advanced computational techniques. This new approach stands to enhance the efficiency and accessibility of protein design, paving the way for groundbreaking advancements in medical and agricultural biotechnology.