A neural-mechanistic hybrid approach improving the predictive power of genome-scale metabolic models

Abstract

Metabolic networks have largely been exploited as mechanistic tools to predict the behavior of microorganisms with a defined genotype in different environments. However, flux predictions by constraint-based modeling approaches are limited in quality unless labor-intensive experiments including the measurement of media intake fluxes, are performed. Using machine learning instead of an optimization of biomass flux - on which most existing constraint-based methods are based - provides ways to improve flux and growth rate predictions. In this paper, we show how Recurrent Neural Networks can surrogate constraint-based modeling and make metabolic networks suitable for backpropagation and consequently be used as an architecture for machine learning. We refer to our hybrid - mechanistic and neural network - models as Artificial Metabolic Networks (AMN). We showcase AMN and illustrate its performance with an experimental dataset of Escherichia coli growth rates in 73 different media compositions. We reach a regression coefficient of R2=0.78 on cross-validation sets. We expect AMNs to provide easier discovery of metabolic insights and prompt new biotechnological applications.

Machine learning in bioprocess development: from promise to practice

“…Finally, a recent preprint introduced Artificial Metabolic Networks (AMN) [54], a concept where fluxes are predicted with a recurrent neural network (RNN) [Box 3]. Here, FBA predictions are used to train the AMN, which can in turn replace the genome-scale model in the application phase.…”

Metabolic Engineering: Methodologies and Applications

“…The AMN model was first trained on simulated FBA data sets and then used to improve predictions of MFA on growth rate data sets in E. coli . This construction allowed for the uncovering of regulation between growth media and the steady-state metabolic phenotype . GEMs and kinetic models have become more popular as engineering tools, but their predictive performance is still insufficient.…”

“…This construction allowed for the uncovering of regulation between growth media and the steady-state metabolic phenotype. 448 GEMs and kinetic models have become more popular as engineering tools, but their predictive performance is still insufficient. We expect to see an increased effort in building new hybrid models that combine mechanistic models and ML models to achieve both high predictive performance and interpretability.…”

Deep Learning Concepts and Applications for Synthetic Biology

“…Following up on this, Faure et al 102 recently showed that artificial metabolic networks can be used to implement RNNs that can be trained to predict growth rates or the consensual metabolic behavior of an organism in response to its environment. As the proposed artificial metabolic networks can optimize various objective functions, they could be used to obtain optimal solutions in various industrial applications, such as searching for the optimum media for the bioproduction of compounds of interest or to engineer microorganism-based decision-making devices for the multiplexed detection of metabolic biomarker or environmental pollutants.…”