AI tools struggle to predict drug immune risks before trials

A workshop at EMBL-EBI explored how machine learning and artificial intelligence could help assess immune risks for new drugs before they enter human trials. Experts gathered to discuss opportunities for these technologies in preclinical risk assessment. However, the group found that predicting the impact of immunogenicity before starting clinical trials is currently impossible. This challenge stems from the difficulty of harmonizing preclinical risk assessment assays with the clinical measurements of anti-drug antibodies. Additionally, machine learning and other AI techniques require large data sets that have been acquired through consistent methods, which are often unavailable. The data available today is often imperfect, further limiting the ability of these tools to make accurate predictions. Industry workflows are currently aligned on the application of these tools, but they recognize that gaps need to be filled with additional data and assays. Readers should understand that while AI offers potential, it cannot yet reliably forecast immune responses in humans based on preclinical data alone.