What will be the first AI-designed drug? These disease-fighting antibodies are top contenders
Antibodies — immune proteins that recognize foreign molecules, such as those made by pathogens, with exquisite specificity — have been a challenge for AI to design. AI models such as AlphaFold have struggled to predict the shape of flexible loop regions of antibodies, which they use to recognize their targets.But new tools developed in the past year — including an updated version of AlphaFold — have proved better at modelling these flexible regions, says Gabriele Corso, a machine-learning scientist at the Massachusetts Institute of Technology in Cambridge. Progress in antibody design has followed.
In October, Corso and his colleagues described the BoltzGen model in a preprint, showing that it can adroitly design ‘nanobodies’ — small, simple antibodies resembling molecules made by sharks and camels — against proteins implicated in cancer, viral and bacterial infections and other diseases. In most cases, the researchers identified antibodies with strong target binding after expressing just 15 of the most-promising designs in cells and testing them in laboratory experiments. However, the molecules were not tested in disease models.
