Exploratory learning in biological cells

The capacity of cells and organisms to respond repeatably to challenging conditions is limited by a finite repertoire of pathways. Beyond this capacity, novel and unforeseen challenges may elicit exploratory dynamics, improvisational in nature, potentially providing adaptation to a much broader array of conditions. Exploratory adaptation, its dynamics and convergence properties are not well understood.

Such phenomena are naturally described as learning processes: Learning entails self-modification of a system under closed-loop dynamics with its environment. In particular, the interactions between system elements are modified - like synapses during learning in the brain, that alter the connections between neurons. Inspired by classic paradigms in Neuroscience, I will describe a theoretic framework for a primitive form of learning that takes place within the single cell. This “exploratory learning” is a random search driven and guided by stress through global feedback. We find this to be a feasible mechanism, but its convergence in high-dimensional gene expression space is non-universal and depends on network properties.

Successfully adapting network ensembles are heterogeneous and have outgoing hubs – the analog of “master regulators” in gene regulatory networks. The role of these hubs in guiding the search process is understood by a mapping to a simpler problem that can be analyzed by mean-field methods and relates to a chaos-suppression phase transition. Results of this theory connect to several experimental observations in cellular systems.

References
H. Schreier, Y. Soen and N. Brenner, "Exploratory adaptation in large random networks", Nature Comm. (2017).
A. Rivkind, H. Schreier, N. Brenner and O. Barak, “Scale free topology as an effective feedback system”. PLoS Comp. Biol. (2020).
A. Shomar, O. Barak and N. Brenner, “Cancer progression as a learning process”, iScience (2022).