AI-Driven Universal Strategy Enhances Protein Engineering Efficiency
A research team led by Prof. GAO Caixia at the Institute of Genetics and Developmental Biology (IGDB), Chinese Academy of Sciences, has introduced an innovative method called AI-informed Constraints for protein Engineering (AiCE). This approach accelerates protein evolution by integrating structural and evolutionary constraints into a generic inverse folding model, eliminating the need for training specialized AI models.
Published in Cell on July 7, 2025, the study addresses key limitations of traditional protein engineering, such as high cost, low efficiency, and poor scalability. Existing AI-based methods often require heavy computational resources, limiting accessibility. AiCE offers a more practical solution that retains predictive accuracy while being broadly usable.
AiCE Single: Precision in Predicting Single Amino Acid Substitutions
The researchers developed AiCE single, a module that predicts high-fitness single amino acid substitutions with improved accuracy. It achieves this by extensively sampling inverse folding models—AI tools that generate amino acid sequences compatible with 3D protein structures—while applying structural constraints.
Benchmarking across 60 deep mutational scanning (DMS) datasets showed AiCE single outperforms other AI-based approaches by 36–90%. The module is effective not only for complex proteins but also for protein–nucleic acid complexes. Incorporation of structural constraints alone improved prediction accuracy by 37%.
AiCE Multi: Tackling Combinatorial Mutations
To handle negative epistatic interactions in multiple mutations, the team introduced AiCE multi. This module integrates evolutionary coupling constraints, enabling accurate predictions of multiple high-fitness mutations while maintaining low computational costs. This significantly broadens AiCE’s applicability for engineering proteins with complex mutation patterns.
Successful Application Across Diverse Proteins
Using AiCE, the researchers evolved eight proteins with varied structures and functions, including deaminases, nuclear localization sequences, nucleases, and reverse transcriptases. These engineered proteins contributed to developing advanced base editors used in precision medicine and molecular breeding.
- enABE8e: A cytosine base editor with approximately 50% narrower editing window.
- enSdd6-CBE: An adenine base editor exhibiting 1.3× higher fidelity.
- enDdd1-DdCBE: A mitochondrial base editor with a 13× increase in activity.
Practical Implications and Future Directions
AiCE provides a straightforward, efficient, and widely applicable framework for protein engineering. By leveraging existing AI inverse folding models with added constraints, it reduces computational demands and enhances interpretability in protein redesign.
This method opens new avenues for researchers aiming to engineer proteins more rapidly and accurately without extensive AI model training. Such advancements could impact fields ranging from therapeutic development to agricultural biotechnology.
For more on AI applications in science and research, explore Complete AI Training's latest AI courses.
Reference
Article: Enhanced protein evolution with inverse folding models using structural and evolutionary constraints
Journal: Cell
DOI: 10.1016/j.cell.2025.06.014
Publication Date: 7-Jul-2025
Your membership also unlocks:
