Sberbank and AIRI Develop Tool to Test AI in Antibody Analysis
Sberbank's Center for Practical Artificial Intelligence and the AIRI Institute have developed SAASBench, a tool that tests how accurately AI models can identify an antibody's intended target among similar proteins.
Antibodies are immune-system proteins that recognize molecules known as antigens. They are used to develop drugs to treat cancer, autoimmune diseases and infections.
“We decided to test AI against a practical question: Can it identify the one protein an antibody is supposed to act on among several similar proteins? We developed SAASBench to compare AI models at the early stages of antibody development. It does not replace the laboratory, but it helps screen out unreliable computational approaches,” said Sergey Ryabov, Sberbank's director of AI transformation.
The tool also tests specificity – an antibody's ability to bind to the intended antigen. As Dmitry Umerenkov, a senior research scientist at AIRI's Strong AI in Medicine Laboratory, explained, an antibody that binds to unintended proteins can reduce diagnostic accuracy and the effectiveness of treatment.
