快猫短视频

Artificial intelligence could develop unique antibody treatments against some forms of cancer, suggesting we could fast-track the development of new immunotherapies.

Antibodies are proteins found in our bodies that can bind to other proteins on pathogens or cancer cells. Some directly neutralise the threat, while others tag the pathogen or cancer cell so the immune system can locate and attack it.

Some antibodies have been turned into drugs that are used to treat people with some forms of cancer, but at present it can take weeks or even months to design and generate antibodies that target specific cancer cells. Therefore, many researchers have begun testing whether AI can speed up the process. But while computer simulations suggest these AI-generated antibodies can bind to desired protein targets, this binding performance has rarely been tested in lab experiments.

Now, at Absci Corporation in New York and his colleagues have used an AI to design new versions of an antibody called trastuzumab that is used to treat breast cancer. To do this, the team fed the AI the structure of the cancer protein of interest 鈥� called human epidermal growth factor 2 (HER2) 鈥� along with the partially complete structure of trastuzumab. The structure given to the AI lacked a critical region of the antibody, called HDR3, that binds to HER2.

Over about a week, the AI generated from scratch 440,000 unique versions of the HDR3 region. In lab experiments, Meier鈥檚 team built 421 of the most promising designs and found they could all bind to the HER2 cancer protein.

What鈥檚 more, the researchers found that three of the AI-generated antibodies bound to HER2 more strongly than trastuzumab. This suggests people could be given lower doses of these new versions of the antibody to achieve the same therapeutic effects, which would save on cost and materials. However, antibodies that bind to targets more strongly , so further cell and animal studies would be needed to confirm this idea.

If novel antibodies are too dissimilar from proteins normally found in the body they could provoke an immune response that damages healthy tissue. The team found the AI-generated antibodies were unlikely to cause such an autoimmune response, by using an AI to compare the structure of the AI-generated antibodies with those normally found in the body.

鈥淚t鈥檚 great to see the researchers having some success in screening their AI-generated antibody libraries,鈥� says at Johns Hopkins University in Baltimore, Maryland. 鈥淎lthough their success rate is small, I鈥檓 glad that they are testing their designs experimentally. There is incredible progress right now in AI approaches to antibody generation.鈥�