Conference Agenda

The use of metabolic networks is extremely valuable for design and optimisation of bioprocesses as they provide great insight into cellular metabolism. In model-based bioprocess optimisation, they have been used successfully, enabling better (economic) objective performance through more accurate network-based models. One of the drawbacks of using a metabolic networks is their underdeterminacy, leading to non-unique flux distributions for a given set of measurements. Flux Balance Analysis (FBA) overcomes this issue by assuming that the cell is trying to fulfill a certain objective function. However, for metabolic networks of higher complexity, FBA can still have non-unique solutions to the LP [1]. Metabolic network reduction can greatly reduce this effect but can be difficult when data is limited.

Structural analysis of the metabolic network through Elementary Flux Modes or Extreme Pathways can help find relevant information in the network. Recently, [2] defined a selection procedure for EFMs to find macroscopic relationships between metabolites. This work expands on this selection concept, presenting a network reduction approach based on the active EPs for a given set of measurements. Many of the pathways present in the network will not be active during the process and a significantly smaller network can therefore be constructed, reducing the underdeterminacy significantly.

The novel approach to network reduction is then applied to a case study of oxygen-limited Escherichia coli. The vast set of EPs is generated for the metabolic network of E. coli. From this set, the most informative EPs are selected based on in-silico data and a smaller network is constructed using only the reactions active in those EPs. This leads to much lower complexity metabolic networks while keeping the necessary information on cellular metabolism for the given process.

References

[1] Mahadevan, R., & Schilling, C. H. (2003). The effects of alternate optimal solutions in constraint-based genome-scale metabolic models. Metabolic engineering, 5(4), 264-276.

[2] Maton, M., Bogaerts, P., & Wouwer, A. V. (2022). A systematic elementary flux mode selection procedure for deriving macroscopic bioreaction models from metabolic networks. Journal of Process Control, 118, 170-184.

Applications of machine learning algorithms in Chemistry and Biology have inspired numerous vector embeddings for biological entities (metabolites, proteins, genes, enzymes, etc.) but due to their different specialisations often disregard contextual information. Disease and biosynthesis pathways are elucidated in increasingly complex ways covering various types of omics data and intricate sequential signalling and reaction pathways. We propose an embedding that relates sequential and multi-modal measurements. Thinking of modalities as directions in space instead of unrelated entities of different types can give a more unified idea of what a molecule or a gene is. Just like the famous example from natural language processing where the vector of king – man + woman = queen we conclude that there is a direction representing gender we can think of a gene and a protein in similar way where a gene implies something about a protein and vice versa. By using a model architecture inspired by natural language processing a pathway sequence is broken down into context pairs of the same or different omics. Subsequently, the model aligns pathway steps in close proximity. Because the embeddings are produced using pathway sequences they can be used to optimise reaction sequences (retrosynthesis) in microbiome or even in human health. In these setting numerous competing pathways interact and can support or deprive each other of substrates. A model that optimises these processes has to respect the relationships between pathways and compounds within different pathways. Therefore, an embedding that makes it easy to infer interacting proteins, genes in molecules that all reside in the same space can achieve a higher performance compared with other embeddings. Our model aims to extend the way flux balance works by transferring it to a ML-based environment where gene knockouts but also inserts and novel pathways are predicted computationally and don’t necessarily rely on a pre-existing characterisation of a strains pathways. Due to the connection to FBA we can use validation of FBA results to validate our model. When metabolites, genes, proteins etc. align closely then we expect them to co-react in similar ways to knockouts. That is when we perturb the system similarly reacting entities should be aligned as they must have a connection through a pathway. Through benchmark classifiers we can compare different embeddings in their ability to identify pathways, predict co-expressions and find targets for optimising pathways or identifying drug targets. We find that the our embedding performs better in these tasks and can thus lie the groundwork in elucidating new unknown pathways, optimising product formation and identifying interactions in disease.

In recent years, the burgeoning interest across various sectors in products derived from microbial production has significantly propelled the evolution of the field of metabolic engineering. This discipline aspires to augment the production of specific target compounds as much as possible through the optimal designs for microbial cell factories[1]. Escherichia coli, Saccharomyces cerevisiae (commonly employed for ethanol production), and Corynebacterium glutamicum (frequently used for amino acid production) stand as the most prevalent biological hosts for constructing these cellular factories. Vibrio natriegens is a Gram-negative bacterium known for its remarkable growth rate, holding promise as a prospective standard biotechnological host for laboratory and industrial bio-production, specifically tailored to produce target metabolites[2].

A genome-scale metabolic model (GSMM) is a cellular model constructed using mathematical methods, encompassing known metabolites within the cell, enzymatic reactions between metabolites, and the genes that express enzymes within the cell. Widely utilized in the field of metabolic engineering, GSMMs are instrumental in the computational, simulation, and analysis of cellular behavior under different gene expression conditions and environmental variations, predicting cell growth and specific metabolite production rates, and serving as essential tools for gene necessity analysis and cell viability prediction. With the flourishing development of genome sequencing technologies, the quantity and quality of available GSMMs have been steadily increasing. In 2023, Coppens et al.[2] developed the first GSMM for Vibrio natriegens, a model that showed good consistency with experimental data. Given the progress in the field of bioinformatics, and considering the efficiency and effectiveness of metaheuristic algorithms in achieving global optimal solutions, scientists have begun to employ metaheuristic algorithms in the analysis of GSMMs for hosts such as Escherichia coli[3].

In this study, we combine different metaheuristic algorithms such as particle swarm optimization (PSO) with the GSMM of Vibrio natriegens, using metaheuristic algorithms to explore optimal gene knockout strategies that can achieve the maximum production flux of specific metabolites. The solution of GSMM is based on analysis methods such as flux balance analysis (FBA) and minimization of metabolic adjustment (MOMA). Simulation results demonstrate that the hybrid approach proposed in this study effectively enhances the production capacity of specific target metabolites in Vibrio natriegens, offering strategic guidance for gene knockouts in practical experimental testing.

References:

[1] Bai L, You Q, Zhang C, Sun J, Liu L, et al. 2023. Systems Microbiology and Biomanufacturing 3: 193-206.

[2] Coppens L, Tschirhart T, Leary DH, Colston SM, Compton JR, et al. 2023. Molecular Systems Biology 19: e10523.

[3] Lee MK, Mohamad MS, Choon YW, Mohd Daud K, Nasarudin NA, et al. 2020. A Hybrid of Particle Swarm Optimization and Minimization of Metabolic Adjustment for Ethanol Production of Escherichia Coli, Cham.