Computational tools for guided discovery and engineering of metabolic pathways

Enzyme Promiscuity Prediction Using Hierarchy-Informed Multi-Label Classification

MOTIVATION

As experimental efforts are costly and time consuming, computational characterization of enzyme capabilities is an attractive alternative. We present and evaluate several machine-learning models to predict which of 983 distinct enzymes, as defined via the Enzyme Commission (EC) numbers, are likely to interact with a given query molecule. Our data consists of enzyme-substrate interactions from the BRENDA database. Some interactions are attributed to natural selection and involve the enzyme's natural substrates. The majority of the interactions however involve non-natural substrates, thus reflecting promiscuous enzymatic activities.

RESULTS

We frame this "enzyme promiscuity prediction" problem as a multi-label classification task. We maximally utilize inhibitor and unlabelled data to train prediction models that can take advantage of known hierarchical relationships between enzyme classes. We report that a hierarchical multi-label neural network, EPP-HMCNF, is the best model for solving this problem, outperforming k-nearest neighbours similarity-based and other machine learning models. We show that inhibitor information during training consistently improves predictive power, particularly for EPP-HMCNF. We also show that all promiscuity prediction models perform worse under a realistic data split when compared to a random data split, and when evaluating performance on non-natural substrates compared to natural substrates.

AVAILABILITY AND IMPLEMENTATION

We provide Python code for EPP-HMCNF and other models in a repository termed EPP (Enzyme Promiscuity Prediction) at https://github.com/hassounlab/EPP.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Integrating thermodynamic and enzymatic constraints into genome-scale metabolic models

Stoichiometric genome-scale metabolic network models (GEMs) have been widely used to predict metabolic phenotypes. In addition to stoichiometric ratios, other constraints such as enzyme availability and thermodynamic feasibility can also limit the phenotype solution space. Extended GEM models considering either enzymatic or thermodynamic constraints have been shown to improve prediction accuracy. In this paper, we propose a novel method that integrates both enzymatic and thermodynamic constraints in a single Pyomo modeling framework (ETGEMs). We applied this method to construct the EcoETM (E. coli metabolic model with enzymatic and thermodynamic constraints). Using this model, we calculated the optimal pathways for cellular growth and the production of 22 metabolites. When comparing the results with those of iML1515 and models with one of the two constraints, we observed that many thermodynamically unfavorable and/or high enzyme cost pathways were excluded from EcoETM. For example, the synthesis pathway of carbamoyl-phosphate (Cbp) from iML1515 is both thermodynamically unfavorable and enzymatically costly. After introducing the new constraints, the production pathways and yields of several Cbp-derived products (e.g. L-arginine, orotate) calculated using EcoETM were more realistic. The results of this study demonstrate the great application potential of metabolic models with multiple constraints for pathway analysis and phenotype prediction.

Designing overall stoichiometric conversions and intervening metabolic reactions

Existing computational tools for de novo metabolic pathway assembly, either based on mixed integer linear programming techniques or graph-search applications, generally only find linear pathways connecting the source to the target metabolite. The overall stoichiometry of conversion along with alternate co-reactant (or co-product) combinations is not part of the pathway design. Therefore, global carbon and energy efficiency is in essence fixed with no opportunities to identify more efficient routes for recycling carbon flux closer to the thermodynamic limit. Here, we introduce a two-stage computational procedure that both identifies the optimum overall stoichiometry (i.e., optStoic) and selects for (non-)native reactions (i.e., minRxn/minFlux) that maximize carbon, energy or price efficiency while satisfying thermodynamic feasibility requirements. Implementation for recent pathway design studies identified non-intuitive designs with improved efficiencies. Specifically, multiple alternatives for non-oxidative glycolysis are generated and non-intuitive ways of co-utilizing carbon dioxide with methanol are revealed for the production of C₂₊ metabolites with higher carbon efficiency.

Multi -omics and metabolic modelling pipelines: challenges and tools for systems microbiology

Integrated -omics approaches are quickly spreading across microbiology research labs, leading to (i) the possibility of detecting previously hidden features of microbial cells like multi-scale spatial organization and (ii) tracing molecular components across multiple cellular functional states. This promises to reduce the knowledge gap between genotype and phenotype and poses new challenges for computational microbiologists. We underline how the capability to unravel the complexity of microbial life will strongly depend on the integration of the huge and diverse amount of information that can be derived today from -omics experiments. In this work, we present opportunities and challenges of multi -omics data integration in current systems biology pipelines. We here discuss which layers of biological information are important for biotechnological and clinical purposes, with a special focus on bacterial metabolism and modelling procedures. A general review of the most recent computational tools for performing large-scale datasets integration is also presented, together with a possible framework to guide the design of systems biology experiments by microbiologists.

Genome-scale metabolic network reconstruction

Bacterial metabolism is an important source of novel products/processes for everyday life and strong efforts are being undertaken to discover and exploit new usable substances of microbial origin. Computational modeling and in silico simulations are powerful tools in this context since they allow the exploration and a deeper understanding of bacterial metabolic circuits. Many approaches exist to quantitatively simulate chemical reaction fluxes within the whole microbial metabolism and, regardless of the technique of choice, metabolic model reconstruction is the first step in every modeling pipeline. Reconstructing a metabolic network consists in drafting the list of the biochemical reactions that an organism can carry out together with information on cellular boundaries, a biomass assembly reaction, and exchange fluxes with the external environment. Building up models able to represent the different functional cellular states is universally recognized as a tricky task that requires intensive manual effort and much additional information besides genome sequence. In this chapter we present a general protocol for metabolic reconstruction in bacteria and the main challenges encountered during this process.

Computational evaluation of factors governing catalytic 2-keto acid decarboxylation

Recent advances in computational approaches for creating pathways for novel biochemical reactions has motivated the development of approaches for identifying enzyme-substrate pairs that are attractive candidates for effecting catalysis. We present an improved structural-based strategy to probe and study enzyme-substrate binding based on binding geometry, energy, and molecule characteristics, which allows for in silico screening of structural features that imbue higher catalytic potential with specific substrates. The strategy is demonstrated using 2-keto acid decarboxylation with various pairs of 2-keto acids and enzymes. We show that this approach fitted experimental values for a wide range of 2-keto acid decarboxylases for different 2-keto acid substrates. In addition, we show that the structure-based methods can be used to select specific enzymes that may be promising candidates to catalyze decarboxylation of certain 2-keto acids. The key features and principles of the candidate enzymes evaluated by the strategy can be used to design novel biosynthesis pathways, to guide enzymatic mutation or to guide biomimetic catalyst design.