Physics-Informed Machine Learning (PIML) reconciled knowledge- and data-driven modeling of dynamical systems, resulting in a number of successful applications, like improving and speeding up climate models simulations. This combination is also critical for advancing Biology, a field famous for grappling with unsettled knowledge and heterogeneous data, for instance when predicting vaccine response to a new virus in a personalized manner. These features render PIML's direct application to biological systems inherently flawed, preventing its full transformative potential. Thus, I propose a paradigm shift toward a new field, Biology-Informed Machine Learning (BIML), where the emphasis will be put on uncertainty quantification, in a broad sense. This will yield machine learning models equipped with a precise sense of ``what they don't know''. Such models will then be applied to some of the most challenging but impactful healthcare problems that are vaccine and cancer research.

2026

2030