\u201cThis is part of a big transformation in bioengineering of going from random library selection methods and evolution methods to rational design,\u201d Baker said.\u00a0<\/p>\n
Conventionally, discovering antibodies is expensive, laborious and time-consuming, requiring the use of animal immunisation tests and extensive screening. <\/p>\n
Scientists wait months for the test animal to produce an antibody response, and the work involves lots of trial and error. Using an AI tool could speed up this process to weeks, without the need for animal testing, and could help scientists find more precise targets.\u00a0<\/p>\n
The researcher\u2019s AI model \u2014 dubbed RFantibody \u2014 is based on a previous model by the same lab that designs new proteins, and is fine-tuned with additional data on antibodies. <\/p>\n
It was trained to generate new antibodies and predict which ones should be tested further in a lab, said Joe Watson, a researcher at the University of Washington and co-founder of Xaira Therapeutics. <\/p>\n
In the study, the team were able to design antibodies that successfully bound to an actual cancer protein. Finding antibodies for cancer is hard because often the difference between a tumour cell and a normal one might be a single protein.\u00a0<\/p>\n
\u201cBecause we know where a piece is, we can just click on that, and tell the model, \u2018this is where I want an antibody that binds,\u2019\u201d said Watson. \u201cThat\u2019s why we think this is the future.\u201d\u00a0\u00a0<\/p>\n
Recommended<\/p>\n
![\"FT](\"https:\/\/www.ft.com\/__origami\/service\/image\/v2\/images\/raw\/https%3A%2F%2Fimages.ft.com%2Fv3%2Fimage%2Fraw%2Fhttps%253A%252F%252Fd1e00ek4ebabms.cloudfront.net%252Fproduction%252F364c2975-478d-4420-9b74-26b562eb4532.jpg%3Fsource%3Dnext-article%26fit%3Dscale-down%26quality%3Dhighest%26width%3D700%26dpr%3D1?source=next&fit=scale-down&dpr=2&width=240\")
<\/a><\/p>\n