- A language model AI created proteins as good as those refined over a million years of evolution.
- Salesforce’s ProGen has developed sequences based on the “sets” of biological proteins.
- Scientists are studying whether AI could identify a treatment for conditions like rheumatoid arthritis and multiple sclerosis.
Artificial intelligence is a master of imitation. Every time scientists design AI — whether to mimic human speech or to master a game like chess — it doesn’t either matches or far exceeds the capabilities of its biological creators. Now, AI has proven that it can even master the art of biology itself.
Researchers at the University of California-San Francisco, the University of California-Berkeley and Salesforce Research, a scientific arm of the SF-based software company, have developed an AI capable of copying evolution itself. That doesn’t mean the AI created some sort of evolutionarily superior superhuman (yet), but instead the AI designed sequences of 20 amino acids that make up proteins. Compared to nature’s manual labor, some of the sequences worked just as well as those generated over millions of years of evolution. The researchers published their findings in the diary natural biotechnology.
Interestingly, scientists didn’t design an AI from scratch, but repurposed one from an unlikely field: a language model. Leveraging Salesforce’s ProGen natural language processing capabilities, the researchers focused on the “sentences” of biological proteins — essentially a language of amino acids.
“Just as words are individually strung together to form sentences of text, amino acids are individually strung together to form proteins,” says Nikhil Naik, Director of AI Research at Salesforce Research. told motherboard. “Building on this finding, we apply neural language modeling to proteins to generate realistic but novel protein sequences.”
After training ProGen on 280 million proteins, the AI was “iteratively optimized by learning to predict the probability of the next amino acid based on past amino acids in a raw sequence,” according to the paper. The team eventually focused on five specific artificial proteins and compared them to an enzyme found in chicken eggs called “chicken egg white lysozyme” – two of the AI-generated proteins scored positively in the comparison.
Overall, Salesforce estimated that 73 percent of ProGen’s proteins could function, compared to 59% of natural proteins, noting that the AI was also able to recognize evolutionary patterns (although it wasn’t specifically designed to do so) . AI designed human proteins before, but this is the first time a language model AI has been able to accomplish this feat.
But the team isn’t interested in just answering the question of whether language model AIs can design proteins. With proteins forming the basis of many diseases, the Salesforce AI Research team is already investigating how ProGen could identify a treatment for conditions like rheumatoid arthritis and multiple sclerosis.
So while some AIs are trained to beat humans at their own game (literally), language models like ProGen could one day eclipse evolution itself and help humans fight some of the world’s weakest health problems.
Darren lives in Portland, has a cat, and writes/edits about science fiction and how our world works. You can find his previous stuff at Gizmodo and Paste if you look hard enough.
