Deep learning dreams up new protein structures

Just as convincing images of cats can be created using artificial intelligence, new proteins can now be made using similar tools. In a report in Nature, a team of researchers describe the development of a neural network that “hallucinates” proteins with new, stable structures.

“The potential to hallucinate brand-new proteins that bind particular biomolecules or form desired enzymatic active sites is very exciting,” said the senior author.

Proteins are string-like molecules found in every cell that spontaneously fold into intricate three-dimensional shapes. These folded shapes are key to nearly every process in biology, including cellular development, DNA repair, and metabolism. But the complexity of protein shapes makes them difficult to study. Biochemists often use computers to predict how protein strings, or sequences, might fold. In recent years, artificial intelligence techniques like neural networks and deep learning have revolutionized the accuracy of this work.

“For this project, we made up completely random protein sequences and introduced mutations into them until our neural network predicted that they would fold into stable structures,” said co-lead author. “At no point did we guide the software toward a particular outcome — these new proteins are just what a computer dreams up.”

In the future, the team believes it should be possible to steer the artificial intelligence so that it generates new proteins with useful features. “We’d like to use deep learning to design proteins with function, including protein-based drugs, enzymes, you name it,” said co-lead author.

The research team generated 2,000 new protein sequences that were predicted to fold. Over 100 of these were produced in the laboratory and studied. Detailed analysis on three such proteins confirmed that the shapes predicted by the computer were indeed realized in the lab.

The senior author notes, “The hallucination approach builds on earlier observations we made together with the Baker lab revealing that protein structure prediction with deep learning can be quite accurate even for a single protein sequence, without recourse to contact predictions usually obtained by analysis of many evolutionary-related protein sequences.”  

“This approach greatly simplifies protein design,” said senior author. “Before, to create a new protein with a particular shape, people first carefully studied related structures in nature to come up with a set of rules that were then applied in the design process. New sets of rules were needed for each new type of fold. Here, by using a deep-learning network that already captures general principles of protein structure, we eliminate the need for fold-specific rules and open up the possibility of focusing on just the functional parts of a protein directly.” 

“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 the senior author.