Learning protein function through autonomous experimental interaction
Biological AI learns primarily from existing observations, but many questions cannot be answered from available data alone. Here we show that AI can instead acquire knowledge by acting directly on biological systems and learning from the consequences. We developed a closed-loop framework in which autonomous agents design protein variants, construct and characterize them in a robotic laboratory,…
Biological artificial intelligence (AI) can learn about protein function by directly interacting with biological systems, according to recent research. This study demonstrates that AI agents can autonomously design protein variants, construct and analyze them in a robotic laboratory, and learn from the results without human intervention. Over the course of about a month, multiple agents independently explored protein sequence space, learning from shared experimental experience.
Applied to glycoside hydrolases, the AI agents discovered enzymes with significantly altered substrate specificity towards non-native sugars. As the agents progressed, they learned the structure of the underlying sequence-function landscape. This experimental experience also revealed determinants of substrate specificity and protein expression that were not initially specified as learning objectives.
The results show that AI can operate continuously in a biological setting to gain knowledge through experience, establishing a framework for biological discovery driven by ongoing experimental interaction.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
