Individuals (and other animals) can be seen as probabilistic models of their world - an embodied model that guides perception and action. Much work on the biological realizations of probabilistic inference has focused on neural circuits, but single-celled organisms also face dynamic and uncertain environments that demand behaviors often couched in cognitive terms such as decision-making and inference. Indeed, ideas from the cognitive sciences have been applied to the biology of individual cells and microbial populations.
For a theory-oriented computational biologist, this raises the question of how a molecular milieu of the type found in signaling systems could support probabilistic inference. This is illustrated by an example in which a single-celled organism, such as yeast, estimates the dynamics of its stochastic environment (that which balances between glucose and galactose as a carbon source) and uses these estimates to regulate its metabolism. The model illustrates how representations to support inference in Markov models could be incorporated into cellular circuits by combining a concentration-dependent scheme to encode probabilities with a molecular mechanism for directional counting.
A different, more abstract case comes from Erik Winfree and his group, who have asked what probability distributions can be realized using a version of chemical reaction networks in which reactions specify interconversions between molecules devoid of structure (i.e. atomic types). While the examples can be invented and the strategies for realizing them may seem unrealistic, they nevertheless give the sense that the kinetics and chemistry of mass action provide a rich means of inference, and that cells could one day be regarded as primitive cognitive machines.
