Deep learning-based kcat prediction enables improved enzyme-constrained model reconstruction

Abstract

Enzyme turnover numbers (kcat) are key to understanding cellular metabolism, proteome allocation and physiological diversity, but experimentally measured kcat data are sparse and noisy. Here we provide a deep learning approach (DLKcat) for high-throughput kcat prediction for metabolic enzymes from any organism merely from substrate structures and protein sequences. DLKcat can capture kcat changes for mutated enzymes and identify amino acid residues with a strong impact on kcat values. We applied this approach to predict genome-scale kcat values for more than 300 yeast species. Additionally, we designed a Bayesian pipeline to parameterize enzyme-constrained genome-scale metabolic models from predicted kcat values. The resulting models outperformed the corresponding original enzyme-constrained genome-scale metabolic models from previous pipelines in predicting phenotypes and proteomes, and enabled us to explain phenotypic differences. DLKcat and the enzyme-constrained genome-scale metabolic model construction pipeline are valuable tools to uncover global trends of enzyme kinetics and physiological diversity, and to further elucidate cellular metabolism on a large scale.

Construction and Analysis of an Enzyme‐constrained Metabolic Model of <em>Corynebacterium glutamicum</em>

“…Unfortunately, there is little data on both for C. glutamicum. There are three ways to improve the coverage of enzyme kinetic parameters in the model: 1) directly populate unknown reactions with mean or median values of enzyme kinetic parameters from other reactions [16][17][18]; 2) expand the EC number annotation information of model reactions using EC number prediction tools [54,55]; and 3) directly predict reactions with unknown parameters based on existing enzyme kinetic parameters via machine learning or deep learning approaches [25,56]. In addition, the kinetic data are mainly sourced from the BRENDA and SABIO-RK databases, which mostly collect in vitro measurements that differ somewhat from the in vivo data.…”

“…In the enzyme-constrained models, kcat and molecular weight (MW) of an enzyme set constraints on the fluxes of the reactions catalyzed by that enzyme. Previous studies have made efforts to automatically acquire kcat values from databases and fill missing values using methods like machine learning [25]. In contrast, few studies paid attention to molecular weight.…”

Interrogating the effect of enzyme kinetics on metabolism using differentiable constraint-based models

“…Moreover, the gradient descent algorithm changes the turnover numbers of all the enzymes simultaneously, thus the imputed enzyme concentrations effectively get changed relative to the initial estimates, attenuating dramatic changes in turnover number estimates of single reactions. Thus, given the increasing number of machine learning generated turnover number estimates [14,48], this algorithm is well suited as a refinement step to increase their accuracy.…”

Construction and Analysis of an Enzyme-Constrained Metabolic Model of Corynebacterium glutamicum

“…In the enzyme-constrained models, k cat and molecular weight (MW) of an enzyme set constraints on the fluxes of the reactions catalyzed by that enzyme. Previous studies have made efforts to automatically acquire k cat values from databases and fill missing values using methods like machine learning [ 25 ]. In contrast, few studies have paid attention to molecular weight.…”