MIT Framework Uses Machine Learning to Design Non-Natural Proteins
PottsMPNN moves beyond mimicking nature to engineer structurally feasible proteins with novel sequences.
MIT biologists have developed a machine-learning framework designed to break the reliance on naturally occurring protein sequences. By moving beyond the imitation of native biological structures, the team aims to engineer entirely novel proteins that remain structurally feasible and stable.
The new framework, called PottsMPNN, improves the precision of computational protein design. Rather than simply reproducing sequences found in nature, PottsMPNN incorporates the physical principles of protein structure and stability. This approach allows the system to better model the sequence-energy landscape, ensuring that resulting synthetic proteins can fold and function as intended. The research was led by graduate student Foster Birnbaum and senior author Amy E. Keating, head of the Department of Biology and a professor of Biological Engineering at MIT. The findings were published in the Proceedings of the National Academy of Sciences (PNAS).
The Shift Toward Synthetic Design
For decades, protein engineering has largely been a process of modification, where scientists tweaked existing natural sequences to enhance their properties. While effective, this method limits the scope of discovery to the boundaries of natural evolution. The shift toward designing non-natural sequences represents a fundamental change in synthetic biology, moving from observation to intentional engineering. By utilizing machine learning to navigate the vast possibilities of amino acid combinations, researchers can now explore a design space that nature never touched.
Implications for Biotechnology
Designing proteins that do not occur in nature opens the door to a new class of biological tools. The ability to create novel sequences with specific, engineered functions allows for the development of synthetic enzymes, more effective drugs, and advanced materials. These innovations could solve complex industrial and medical problems that natural biology is unequipped to handle, such as breaking down synthetic pollutants or targeting specific disease markers with unprecedented precision.
Future Directions
As the PottsMPNN framework demonstrates the viability of non-natural sequence design, the next phase of research will likely focus on the practical application of these proteins in vivo. While the framework improves the computational prediction of stability, the real-world performance of these novel proteins in complex biological environments remains the primary frontier. Researchers will now look to verify how these synthetic structures interact with living systems and whether they can be scaled for industrial production.