ChatGPT for Biology: A New AI Whips Up Designer Proteins With Only a Text Prompt
1 min read
Summary
Scientists are using large language models (LLMs) as inspiration for developing protein language models that design novel proteins, with instructions provided in natural language.
One algorithm, called Pinal, which was developed last month by researchers at Westlake University in China, can generate multiple proteins that, when tested in living cells, were capable of breaking down alcohol.
The algorithm links natural language prompts to protein structures, which in turn are translated into protein sequences via another algorithm called SaProt.
Pinal’s ease of use could unleash a new wave of innovation in biotechnology, with its creators claiming it is the first time a functional enzyme has been designed using only natural language.
The technology is far from perfect, however, and the team suggests that, as with current LLMs, the final protein design output is tainted by the ambiguous nature of natural language inputs.
This means that prompts must be phrased carefully in order to generate successful protein sequences.
