Protein design focuses on creating specific antibodies for medical treatments, biosensors for diagnostic purposes, or enzymes to catalyze chemical reactions. A global research team has now introduced a novel technique for designing larger proteins more effectively and producing them in laboratory settings with the desired characteristics. Their innovative approach utilizes the potential of the AI-driven software Alphafold2, which was awarded the Nobel Prize in Chemistry in 2024.
Protein design focuses on creating specific antibodies for medical treatments, biosensors for diagnostic purposes, or enzymes to catalyze chemical reactions. A global research team has now introduced a novel technique for designing larger proteins more effectively and producing them in laboratory settings with the desired characteristics. Their innovative approach utilizes the potential of the AI-driven software Alphafold2, which was awarded the Nobel Prize in Chemistry in 2024.
Proteins serve crucial functions in our bodies—acting as building blocks, transport mechanisms, enzymes, or antibodies. Because of this importance, researchers are actively engaged in recreating proteins or designing entirely new ones that do not exist in nature. These artificial proteins might be tailored to attach to specific viruses or to transport medications, for instance. With advancements in machine learning, scientists are increasingly turning to these technologies for their designs. This year’s Nobel Prize in Chemistry recognized significant advancements in this field, honoring David Baker, a leader in the design of de novo proteins, along with the creators of Alphafold2, Demis Hassabis and John Jumper. The software allows researchers to accurately predict protein structures using computational methods.
A research team led by Hendrik Dietz, a Professor of Biomolecular Nanotechnology at the Technical University of Munich (TUM), and Sergey Ovchinnikov, a Professor of Biology at MIT, has developed a method that enhances Alphafold2’s precise structure predictions through a technique known as gradient descent for streamlined protein design. Their findings have been published in the journal Science.
Gradient descent is a widely used optimization method that adjusts parameters step by step to minimize any deviations from a target function until the ideal result is achieved. In the context of protein design, this technique compares the structures of new proteins predicted by AlphaFold2 with the desired structure. This allows for further fine-tuning of the designed amino acid chains and their configurations, which are crucial for the protein’s stability and functionality, influenced by delicate energetic interactions.
Virtual Superposition of Building Blocks
This new method enhances the design of large proteins and enables tailored customization for specific properties, such as precise binding to other proteins. This design approach marks a departure from previous methodologies in several ways.
“Our design process for new proteins begins by disregarding the conventional limitations of physical possibility. Typically, only one of the 20 available building blocks is considered at each position within the amino acid sequence. However, we implement a variant where all possibilities are virtually layered together,” explains Christopher Frank, a doctoral student in Biomolecular Nanotechnology and the study’s lead author.
This virtual layering of options does not translate directly into a constructible protein, but it facilitates iterative optimization of the protein. “We refine the amino acid arrangement through multiple iterations until the new protein closely resembles the intended structure,” adds Christopher Frank. This refined structure then helps determine the amino acid sequence that can be synthesized into a protein in the laboratory.
The Essential Test: Do Predictions Match Reality?
The crucial challenge for all newly engineered proteins is to determine whether the actual structures align with the predicted designs and intended functions. Through this new approach, the team virtually designed more than 100 proteins, which they synthesized in the lab and tested. “Our results confirmed that the structures we created closely matched those we produced,” shares Christopher Frank.
Using this novel method, they successfully generated proteins made up of as many as 1000 amino acids. “This advancement brings us closer to the size of antibodies, and similar to antibodies, we can incorporate multiple desired functions into these proteins,” Hendrik Dietz explains. “For example, they could include motifs that recognize and neutralize pathogens.”
