In Silico vs In Vivo: Scientists Demand Strict Validation for AI in Immunology

A sobering signal for the medical AI market. On May 6, 2026, researchers at the University of South Florida (USF) published a paper in Nature Machine Intelligence on predicting immune responses using neural networks (such as the PanPep model).

The authors' conclusion is unequivocal: algorithms perform excellently in controlled laboratory conditions (In silico), but transferring them to real clinical practice entails enormous risks. Researchers are calling for the creation of strict standards for validating AI models using real patient biological data. This news continues the trend of testing consumer AI solutions for hallucinations in medicine. Trusting an algorithm to write programming code is one thing, but trusting it to predict the human body's immune response is a task requiring zero tolerance for statistical errors.

Source: University of South Florida / Nature