Actionable Forecasting as a Determinant of Biological Adaptation

Organisms continuously adapt to changing environments to survive. Here, contrary to the prevailing view that predictive strategies are essential for perfect adaptation, it is shown that biological systems can precisely track their optimal state by adapting to a non-anticipatory actionable target that integrates the current optimum with its rate of change. Predictive mechanisms, such as circadian rhythms, are beneficial for accurately inferring the actionable target when environmental sensing is slow or unreliable. A new mathematical framework is developed, showing that dynamics-informed neural networks embodying these principles can efficiently capture biological adaptation even in noisy environments. These results provide fundamental insights into the interplay between forecasting, control, and inference in biological systems, redefining adaptation strategies and guiding the design of advanced adaptive biomolecular circuits.