Just as artificial intelligence can be used to create compelling images of cats, similar tools can now be used to create new proteins. In a report published in Nature, the researchers describe the development of a neural network that “hallucinates” proteins with new, stable structures.
Proteins, which are chain-like molecules found in every cell, spontaneously fold into complex three-dimensional shapes. These folded forms are essential for almost all biological processes, including cell development, DNA repair, and metabolism. But the complexity of protein forms makes them difficult to study. Biochemists often use computers to predict how protein chains, or sequences, might fold. In recent years, deep learning has revolutionized the precision of this work.
“For this project, we invented completely random protein sequences and introduced mutations into them until our neural network predicted that they would fold into stable structures,” said co-author Ivan Anishchenko, acting professor of biochemistry at the University of Washington School of Medicine and researchers in David Baker’s laboratory at the University of Washington’s Institute of Medicine for Protein Design.
At no point did we direct the software to a specific result, said Anishchenko. These new proteins are exactly what a computer imagines.
The team believes that in the future it should be possible to induce artificial intelligence to generate new proteins with useful properties.
We want to use deep learning to design proteins that have function, including protein-based drugs, enzymes, you name it, said co-author Sam Pellock, a postdoctoral fellow at the Baker lab.
The research team, which included scientists from UW Medicine, Harvard University, and Rensselaer Polytechnic Institute (RPI), generated 2,000 new protein sequences that were predicted to fold, more than 100 of which were made and studied in the laboratory. Such proteins confirmed that the shapes predicted by the computer were actually made in the laboratory.
Our nuclear magnetic resonance (NMR) studies, along with the X-ray crystal structures identified by the University of Washington team, demonstrate the remarkable precision of the protein designs generated by the hallucination approach, said co-author Theresa Ramelot, a senior researcher at RPI in Troy, New York.
Gaetano Montelione, co-author and professor of chemistry and chemical biology at the RPI, noted. “The hallucination approach is based on observations we made in collaboration with Baker’s lab, which show that predicting deep learning protein structure can be fairly accurate even for a single protein sequence with no natural relatives. The potential to hallucinate brand new proteins that bind particular biomolecules or form desired enzymatic active sites is very exciting”.
This approach greatly simplifies protein design, said lead author David Baker, professor of biochemistry at the University of Washington School of Medicine, who received the 2021 Breakthrough Award in Life Sciences. To create a new protein with a certain shape, people first carefully examined related structures in nature in order to develop a set of rules that was then applied in the design process. Each new set of rules required new sets of rules.By using a deep learning network that already captures the general principles of protein structure, we eliminate the need for specific folding rules and open up the possibility of focusing directly on the functional parts of a protein.
“Exploring how to best use this strategy for specific applications is now an active area of research, and this is where I expect the next breakthroughs,” said Baker.
