Generative machine learning produces kinetic models that accurately characterize intracellular metabolic states

Abstract

Large omics datasets are nowadays routinely generated to provide insights into cellular processes. Nevertheless, making sense of omics data and determining intracellular metabolic states remains challenging. Kinetic models of metabolism are crucial for integrating and consolidating omics data because they explicitly link metabolite concentrations, metabolic fluxes, and enzyme levels. However, the difficulties in determining kinetic parameters that govern cellular physiology prevent the broader adoption of these models by the research community. We present RENAISSANCE (REconstruction of dyNAmIc models through Stratified Sampling using Artificial Neural networks and Concepts of Evolution strategies), a generative machine learning framework for efficiently parameterizing large-scale kinetic models with dynamic properties matching experimental observations. We showcase RENAISSANCE's capabilities through three applications: generation of kinetic models of E. coli metabolism, characterization of intracellular metabolic states, and assimilation and reconciliation of experimental kinetic data. We provide the open-access code to facilitate experimentalists and modelers applying this framework to diverse metabolic systems and integrating a broad range of available data. We anticipate that the proposed framework will be invaluable for researchers who seek to analyze metabolic variations involving changes in metabolite and enzyme levels and enzyme activity in health and biotechnological studies.

Sensitivity analysis of biochemical systems using bond graphs

“…Thus, determining the ‘stiff’ combinations of parameters can help to optimize for knowledge gained with limited experimental resources. Furthermore, analysing the time constants can be used to assess the biological relevance of parameter sets in machine learning approaches to parametrization [ 80 , 81 ]. Our approach builds on existing techniques by reframing parameters in a thermodynamically consistent context, which can in some cases reduce the number of parameters and better distinguish between the individual contributions of reactions and metabolites [ 82 , 83 ].…”

Resource Allocation in Mammalian Systems

“…Deep learning provides an avenue forward to robustly estimate such parameters from molecular features (Kroll et al, 2023;. Deep learning and other algorithmic approaches may be combined with GEMs to leverage flux predictions and metabolic network topology to estimate enzyme kinetic parameters (Andreozzi et al, 2016;Choudhury et al, 2023;Gopalakrishnan et al, 2020;Heckmann et al, 2020Heckmann et al, , 2018. Finally, measuring multiple biological layers (Figure 1C) is important to fully and explicitly account for the resource costs of biological activity and understand how resource allocation decisions propagate to phenotype.…”

Sensitivity Analysis of Biochemical Systems Using Bond Graphs

“…Thus, determining the 'stiff' combinations of parameters can help to optimise for knowledge gained with limited experimental resources. Furthermore, analysing the time constants can be used to assess the biological relevance of parameter sets in machine learning approaches to parameterisation [75,76]. Our approach builds on existing techniques by reframing parameters in a thermodynamically consistent context, which can in some cases reduce the number of parameters and better distinguish between the individual contributions of reactions and metabolites [77,78].…”