A machine-learning program dreamed up hundreds of structures. Ian C. Haydon is a researcher at the Medicine Institute for Protein Design.
Similar tools can be used to make new proteins, just as they can be used to create convincing images of cats. Researchers describe the development of a neural network that "hallucinates" proteins with new, stable structures.
Every cell has string-like molecule called a proteins that folds into three-dimensional shapes. The folded shapes are key to nearly every biological process. The complexity of the shapes makes them hard to study. Biochemists use computers to predict how a sequence might fold. Deep learning has made this work more accurate.
Ivan Anishchenko is an acting instructor of biochemisty at the University of Washington School of Medicine and a researcher.
"These new proteins are just what a computer dreams up, and we didn't guide the software toward a particular outcome," Anishchenko said.
In the future, the team believes it should be possible to steer the artificial intelligence so that it can generate useful features.
The Baker lab wants to use deep learning to design drugs with function, according to co-lead author Sam Pellock.
The research team, which included scientists from the University of Wisconsin Medicine, Harvard University, and the Rensselaer Polytechnic Institute, generated two thousand newProtein Sequences that were predicted to fold. Over 100 of these were studied. The shapes predicted by the computer were realized in the lab.
Theresa Ramelot, a senior research scientist at RPI in Troy, said that their studies along with X-ray crystal structures determined by the University of Washington show the remarkable accuracy of the designs.
Gaetano Montelione is a professor of chemistry and chemical biology. The Baker lab and I made observations that showed that deep learning can be quite accurate even for a single sequence with no natural relatives. It is very exciting to have the chance to hallucinate brand new proteins that bind specific biomolecules or form desired active sites.
David Baker, a professor of biochemistry at the University of Washington School of Medicine, received a prize for his work on the simplification of the design of genes. To create a newProtein 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. There were new rules needed for each fold. By using a deep-learning network that already captures general principles of the structure, we can eliminate the need for fold-specific rules and focus on just the functional parts of a protein.
The next breakthrough will be in the area of exploring how to best use this strategy for specific applications.
David Baker is the author of De novo protein design by deep network hallucination. www.nature.com/articles/s41586-021-04184-w
Nature journal information.
Deep learning dreams up new structures on December 1st, 2021.
The document is copyrighted. Any fair dealing for the purpose of private study or research cannot be reproduced without written permission. The content is not intended to be used for anything other than information purposes.
