Researchers at the University of Michigan have developed a novel machine learning model, led by Tessier's team, aimed at enhancing the design of antibody-based medicines.
These innovative machine-learning algorithms have the capability to identify problematic regions within antibodies that make them susceptible to binding with non-target molecules.
Furthermore, these algorithms can precisely pinpoint the specific positions in the antibodies responsible for these binding issues and propose modifications to rectify these problems without introducing new challenges.
The machine learning models can be applied to existing antibodies, newly developed ones, and even antibodies that are yet to be synthesized.
Antibodies function by binding to specific molecules called antigens found on disease-causing agents, such as the spike protein on the COVID-19 virus. Once bound, antibodies either directly neutralize the harmful agents or trigger the body's immune cells to do so.
Unfortunately, antibodies that are engineered to bind strongly and rapidly to their specific antigens can also bind to non-antigen molecules. This unintended binding can lead to the removal of antibodies before they can effectively target a disease. Additionally, such antibodies can bind to other antibodies of the same type, forming dense solutions that do not flow smoothly through the needles used to administer antibody drugs.
For an antibody to become a successful drug, it must specifically target its intended antigen, avoid unintended binding, and not stick to other antibodies. However, many clinical-stage antibodies struggle to meet these criteria.
In their recent study, Tessier's team assessed the performance of 80 clinical-stage antibodies in a laboratory setting and discovered that 75 percent of these antibodies exhibited issues, including binding to the wrong molecules and self-interaction.
These innovative machine learning models from the University of Michigan have the potential to revolutionize the development of antibody-based drugs by offering a more efficient and precise method for optimizing antibodies for therapeutic purposes.
