It takes more than a decade to develop a usable vaccine. COVID-19 veered off from this marathon, as scientists ran a sprint to design a vaccine in less than a year. Soon after the COVID tide ebbed, the AI surge began. Last year, biologists successfully created an entirely new antibody molecule using just AI. The only downside: it lacked the potency of commercially available antibody drugs, which account for the lion’s share of the market. After a year of progress, scientists believe they have reached the point at which they can consider transforming AI-designed antibodies into potential therapies. In the last few weeks, many scientists have reported using open-source AI tools to generate antibodies with multiple properties of an antibody drug.
“These latest efforts are compelling advances that enable a democratization of antibody engineering,” says Chang Liu, a synthetic biologist at the University of California, Irvine.
De novo antibody synthesis, or the synthesis of antibodies from scratch, has gained traction recently, and AI can facilitate the rapid processing of these bioactive materials.
Therapeutic antibodies are screened for their interactions and their ability to recognize specific targets, called antigens. Sometimes, these antibodies fail to recognize the right binding region or form a weak interaction.
Major hurdles lie in antibody precision, as many interactions result in imperfect binding or binding to the wrong region. AI-guided antibodies can be custom-made to target specific areas of antigens or of enzymes that drive the disease.
AI has been particularly challenging for antibody design, as it has not yet succeeded in predicting the chemical structures of the flexible loop regions that recognize target sites. An updated version of AlphaFold has proven better at modelling these regions.
Gabriele Corso, a machine-learning scientist at the Massachusetts Institute of Technology in Cambridge, and colleagues used the BoltzGen model to design “nanobodies”, which are simple and small antibodies that sharks and camels produce. In approximately 15 antibody sequences, the scientists identified strong binding to proteins implicated in cancer, viral, and bacterial diseases.
Progress in the design of drug-like antibodies was reported last month by scientists at Nabla and Chai Discovery in San Francisco, California. The designer molecules identified diverse disease targets, such as G protein-coupled receptors (GPCRs), a recognized challenge in antibody design. Custom properties can be added to the produced antibodies, thereby facilitating production.
Clinical trials of AI-designed drugs are not too far off, and Generate Biomedicine, in Somerville, Massachusetts, conducted a clinical trial of an antibody drug used to treat severe asthma. The trial didn’t use any new antibodies, but optimized an existing antibody drug to improve binding, stability, and other properties. Most importantly, designing antibody drugs through AI can help start-ups to crank out capable designs.
The race to an AI-designed antibody drug is entering the final lap. And drug design has never been more exciting.
Copyright @smorescience. All rights reserved. Do not copy, cite, publish, or distribute this content without permission.
