In the past, researchers have altered proteins by cloning or inducing microbial mutations, or manually modifying amino acid sequences to design proteins. These traditional methods are both cumbersome and can result in proteins that are misfolded or not successfully expressed in cells.
Advances in computational protein design and machine learning algorithms have made it possible that tasks such as designing customized proteins, which have a high degree of uncertainty, can now be efficiently accomplished on a computer. An article we cite here discusses the potential of AI in the field of protein design and the major challenges it faces.
The article describes some of the AI tools for predicting protein structure, such as:
The design of new proteins through AI has been widely recognized. In 2024, AlphaFold and its development team were awarded the Nobel Prize in Chemistry for their outstanding achievements in protein structure prediction and design. The award signaled the far-reaching impact of AI in the life sciences, particularly revolutionizing the field of protein design.
Despite the impressive progress made by AI tools in the field of protein design, challenges remain.
One challenge in protein design is predicting how proteins will bind to each other, which is critical for drug development and is key to drug design. Generative AI tools such as RFdiffusion and AlphaProteo have made predicting protein binding relatively simple.
However, in the context of drug-protein binding, AI performance is currently still unsatisfactory, as small molecules often have diverse chemical structures and properties.
Researchers hope to use computational tools to design enzymes with entirely new functions. However, similarity in protein shape does not always imply similarity in function. Understanding the functional connections between enzymes and how to recreate new functions is a major challenge in the use of AI for protein design.
The conformation of proteins is affected by temperature, pH, the chemical environment and binding to other molecules. The dynamic properties of proteins are critical to their function, especially when signaling and catalytic reactions are involved. Computing all possible conformations of proteins is a challenging task, so it is crucial to design AI models that can simulate these dynamic changes.
Proteins can be used not only as enzymes but also as building blocks for self-assembly into complex structures capable of carrying cargo into cells and generating mechanical forces. Through rational design, proteins can be endowed with a variety of novel functions for solving practical biomedical problems.
With AI computational tools, researchers can predict and optimize the process of protein self-assembly to improve its stability and functionality. However, because there are not enough easy-to-understand examples to refer to, for more complex structures, machine learning is currently limited in what it can do.
The authors say that even the best predictive algorithms struggle to create accurate proteins all at once. Algorithms rarely have the opportunity to learn from mistakes because researchers tend not to publish negative results, even though these failures may lead to useful information.
AI algorithms are changing the research and application landscape of protein design. Future research directions for AI tools in the field of protein design include improved algorithms, increased data, and enhanced collaborations, which will make it possible to customize proteins.
Our company provides advanced AI models dedicated to helping researchers predict the 3D structure of proteins with high accuracy, expanding the prospect of protein applications in pharmaceuticals and other fields. If you have any questions about our AI technology, please feel free to contact us for more details.
Original Article:
Reardon S. (2024). Five protein-design questions that still challenge AI. Nature. 2024, 635(8037): 246-248.
Related Services
Our mission is to accelerate the development of life-saving drugs by leveraging cutting-edge AI technologies.
Certification
