Reservoir Computing with Bacteria

Abstract

This study explores the use of bacterial strains in reservoir computing (RC) to solve regression and classification tasks. We employ an Escherichia coli K-12 MG1655 strain as the physical reservoir, training it on M9 minimal media supplemented with 28 metabolites, and measuring growth rates across various media compositions. Our physical RC system, using an Escherichia coli strain, demonstrates superior performance compared to multi-linear regression or support-vector machine and comparable performance to multi-layer perceptron in various regression and classification tasks. Additionally, the performances of RC based on genome-scale metabolic models for several bacterial species correlate with the diversity and complexity of phenotypes they produce. These findings highlight the potential of bacterial RC systems for complex computational tasks typically reserved for digital systems and suggest future research directions, including optimizing feature-to-nutrient mappings and integrating with emerging technologies for enhanced computing capabilities.