Cells are computational entities specialized in statistical learning

We hypothesize that cells are computational entities whose biochemical mechanisms are specialized in statistical learning. That is, cells learn by extracting regularities from their surroundings, predicting changes in environmental conditions, and reorganizing their biochemical networks for optimizing a risk function (e.g., ATP production), thereby preserving viability under dynamically fluctuating environmental conditions. This is not a metaphor, nor a proposal to apply machine-learning techniques for understanding biological data: it is the claim that cellular functioning is a physical realization of a process of statistical learning. Applying machine-learning methods to study cells is a possible methodological consequence of this hypothesis, not the hypothesis itself.