Information Benchmark for Biological Sensors Beyond Steady States -- Mpemba-like sensory withdrawal effect

TL;DR

This paper develops a comprehensive theory to evaluate transient dynamics of biological sensors, revealing a counter-intuitive sensory boost after high ligand exposure followed by resetting, akin to a Mpemba-like effect, with insights from machine learning.

Contribution

It introduces a novel benchmark for transient sensory performance and uncovers a new Mpemba-like effect in multi-state biological sensors, linking sensor structure to performance.

Findings

Counter-intuitive sensory boost after high ligand exposure and resetting.

A new benchmark for assessing transient sensory performance.

Machine learning links sensor structure to performance.

Abstract

Biological sensors rely on the temporal dynamics of ligand concentration for signaling. The sensory performance is bounded by the distinguishability between the sensory state transition dynamics under different environmental protocols. This work presents a comprehensive theory to characterize arbitrary transient sensory dynamics of biological sensors. Here the sensory performance is quantified by the Kullback-Leibler (KL) divergence between the probability distributions of the sensor's stochastic paths. We introduce a novel benchmark to assess a sensor's transient sensory performance arbitrarily far from equilibrium. We identify a counter-intuitive phenomenon in multi-state sensors: while an initial exposure to high ligand concentration may hinder a sensor's sensitivity towards a future concentration up-shift, certain sensors may show a boost in sensitivity if the initial high concentration exposure is followed by a transient resetting at a low concentration environment. The boosted performance exceeds that of a sensor starting from an initially low concentration environment. This effect, reminiscent of a drug withdrawal effect, can be explained by the Markovian dynamics of the multi-state sensor, similar to the Markovian Mpemba effect. Moreover, an exhaustive machine learning study of 4-state sensors reveals a tight connection between the sensor's performance and the structure of the Markovian graph of its states.