A spatial representation of information underlies probabilistic computation in cells

Abstract

Seeking a signal source in unstructured environments is a fundamental challenge in robotics. Similarly, cells in tissues track signal sources using noisy, fragmented molecular gradients, shaped by fluid flow and extracellular matrix interactions. However, the precise algorithm cells use for source seeking is unknown. We show that cells can perform source seeking using a biophysical implementation of a computational algorithm called Bayes filtering. Specifically, the spatial distribution of molecules within the cell encodes a probability distribution over source location, and intracellular transport processes update this distribution. Live-cell imaging and spatial proteomics reveal that receptor dynamics in vivo matches the evolution of belief distributions under Bayes filtering. Unlike standard Bayes filtering, the cellular implementation adapts to fluctuating measurement noise without explicitly estimating noise statistics. When translated to traditional robotics algorithms, this cell-inspired adaptation enables robust navigation without continuously estimating signal statistics. Our results show that cells can leverage spatial organization to implement probabilistic algorithms, bridging cellular behavior and engineered systems.