An artificial intelligence (AI) algorithm initially designed for reading and writing language can now generate protein sequences that work as well or better than natural proteins.
The program, called ProGen, was developed by the software company Salesforce. Researchers trained this algorithm, known as a language model, on a database of 280 million natural protein sequences, and then prompted it to develop protein sequences of its own. They then synthesized a subset of the proteins in the lab to test their activity.
“The surprising thing was that our artificial proteins worked as well, and as frequently, as natural proteins,” says James Fraser, a bioengineer at the Univ. of California, San Francisco, who collaborated on the research. “The previous evolutionarily inspired state-of-the-art [artificial] proteins did not work nearly as well.”
Natural proteins are workhorses for many reactions, but they’re often functional only in a small envelope of temperature or pH, leading to intense interest in designing custom proteins from scratch that can work in more extreme environments than average. Until the advent of machine learning, the most successful efforts toward this goal drew from natural protein evolution,...
Would you like to access the complete CEP News Update?
No problem. You just have to complete the following steps.
You have completed 0 of 2 steps.
-
Log in
You must be logged in to view this content. Log in now.
-
AIChE Membership
You must be an AIChE member to view this article. Join now.
Copyright Permissions
Would you like to reuse content from CEP Magazine? It’s easy to request permission to reuse content. Simply click here to connect instantly to licensing services, where you can choose from a list of options regarding how you would like to reuse the desired content and complete the transaction.
