Neural-like computation using bacterial metabolism to solve machine-learning problems
Abstract
Throughout evolution, bacteria have acquired the ability to sense variations
of the concentrations of nutrients in their growth medium. According to the
medium composition, they adapt their metabolic behaviour by activating or
repressing the appropriate metabolic pathways through a wide array of regulation mechanisms including transcriptional, translational or post-translational
responses. Bacterial metabolism can therefore be compared to an algorithm taking as inputs the media composition and yielding as outputs metabolic fluxes
describing its metabolic phenotype.
Our work focuses on this perspective of the bacterial metabolism as an
information processing unit. Our objective is to demonstrate that E. coli’s
metabolism is capable of neural-like computation and to assess to what extend
it can solve classical machine-learning problems (whether regression or classification).
Our first step has been to generate an accurate model of E. coli’s metabolism,
using the AMN (Artificial Metabolic Network), a metabolic hybrid model previously developed in our lab. This model has then been used to solve machinelearning problems of different complexities in order to assess the capacity of our
metabolic model.
Domains
Quantitative Methods [q-bio.QM]| Origin | Files produced by the author(s) |
|---|