AI can now design proteins from plain-language descriptions

AI can now design proteins from plain-language descriptions

AI can now design proteins from plain-language descriptions

Researchers at Westlake University in China have built an AI system called Pinal that designs candidate proteins from a plain written description of what the protein should do, with the work published in Nature. The system first turns the description into a structural blueprint, then generates a matching amino acid sequence, removing the need for users to specify complex molecular structures. In tests it tackled four jobs, including a glowing protein, a plastic-breaking enzyme and a protein that helps capture carbon, with the best carbon-capture candidate working 1.7 times better than a natural version. The researchers caution that results are still uneven and precise control remains difficult, so more work is needed before it becomes a routine lab tool.

Researchers at Westlake University in China have built an AI system called Pinal that designs candidate proteins from a plain written description of what the protein should do, with the work published in Nature. The system first turns the description into a structural blueprint, then generates a matching amino acid sequence, removing the need for users to specify complex molecular structures. In tests it tackled four jobs, including a glowing protein, a plastic-breaking enzyme and a protein that helps capture carbon, with the best carbon-capture candidate working 1.7 times better than a natural version. The researchers caution that results are still uneven and precise control remains difficult, so more work is needed before it becomes a routine lab tool.

Researchers at Westlake University in China have built an AI system called Pinal that designs candidate proteins from a plain written description of what the protein should do, with the work published in Nature. The system first turns the description into a structural blueprint, then generates a matching amino acid sequence, removing the need for users to specify complex molecular structures. In tests it tackled four jobs, including a glowing protein, a plastic-breaking enzyme and a protein that helps capture carbon, with the best carbon-capture candidate working 1.7 times better than a natural version. The researchers caution that results are still uneven and precise control remains difficult, so more work is needed before it becomes a routine lab tool.