Biophysics Journal Club

Trade-offs between cost and information in cellular prediction

Callum Wareham, SFU Physics
Location: P8445.2

Wednesday, 12 June 2024 11:30AM PDT


Living cells can leverage correlations in environmental fluctuations to  predict the future environment and mount a response ahead of time. To  this end, cells need to encode the past signal into the output of the  intracellular network from which the future input is predicted. Yet,  storing information is costly while not all features of the past signal  are equally informative on the future input signal. Here, we show for  two classes of input signals that cellular networks can reach the  fundamental bound on the predictive information as set by the  information extracted from the past signal: Push–pull networks can reach  this information bound for Markovian signals, while networks that take a  temporal derivative can reach the bound for predicting the future  derivative of non-Markovian signals. However, the bits of past  information that are most informative about the future signal are also  prohibitively costly. As a result, the optimal system that maximizes the  predictive information for a given resource cost is, in general, not at  the information bound. Applying our theory to the chemotaxis network of  Escherichia coli reveals that its adaptive kernel is optimal for  predicting future concentration changes over a broad range of  background concentrations, and that the system has been tailored to  predicting these changes in shallow gradients.

by  Age J. Tjalma, et al.
PNAS 120 (41), e2303078120 (2023).