Fast development of Pichia pastoris GS115 Mut+ cultures employing batch-to-batch control and hybrid semi-parametric modeling

A General Hybrid Modeling Framework for Systems Biology Applications: Combining Mechanistic Knowledge with Deep Neural Networks under the SBML Standard

In this paper, a computational framework is proposed that merges mechanistic modeling with deep neural networks obeying the Systems Biology Markup Language (SBML) standard. Over the last 20 years, the systems biology community has developed a large number of mechanistic models that are currently stored in public databases in SBML. With the proposed framework, existing SBML models may be redesigned into hybrid systems through the incorporation of deep neural networks into the model core, using a freely available python tool. The so-formed hybrid mechanistic/neural network models are trained with a deep learning algorithm based on the adaptive moment estimation method (ADAM), stochastic regularization and semidirect sensitivity equations. The trained hybrid models are encoded in SBML and uploaded in model databases, where they may be further analyzed as regular SBML models. This approach is illustrated with three well-known case studies: the Escherichia coli threonine synthesis model, the P58IPK signal transduction model, and the Yeast glycolytic oscillations model. The proposed framework is expected to greatly facilitate the widespread use of hybrid modeling techniques for systems biology applications.

Knowledge transfer across cell lines using hybrid Gaussian process models with entity embedding vectors

To date, a large number of experiments are performed to develop a biochemical process. The generated data is used only once, to take decisions for development. Could we exploit data of already developed processes to make predictions for a novel process, we could significantly reduce the number of experiments needed. Processes for different products exhibit differences in behaviour, typically only a subset behave similar. Therefore, effective learning on multiple product spanning process data requires a sensible representation of the product identity. We propose to represent the product identity (a categorical feature) by embedding vectors that serve as input to a Gaussian process regression model. We demonstrate how the embedding vectors can be learned from process data and show that they capture an interpretable notion of product similarity. The improvement in performance is compared to traditional one-hot encoding on a simulated cross product learning task. All in all, the proposed method could render possible significant reductions in wet-lab experiments.

Integrating metabolic modeling and population heterogeneity analysis into optimizing recombinant protein production by Komagataella (Pichia) pastoris

The methylotrophic yeast Komagataella (Pichia) pastoris has become one of the most utilized cell factories for the production of recombinant proteins over the last three decades. This success story is linked to its specific physiological traits, i.e., the ability to grow at high cell density in inexpensive culture medium and to secrete proteins at high yield. Exploiting methanol metabolism is at the core of most P. pastoris-based processes but comes with its own challenges. Co-feeding cultures with glycerol/sorbitol and methanol is a promising approach, which can benefit from improved understanding and prediction of metabolic response. The development of profitable processes relies on the construction and selection of efficient producing strains from less efficient ones but also depends on the ability to master the bioreactor process itself. More specifically, how a bioreactor processes could be monitored and controlled to obtain high yield of production. In this review, new perspectives are detailed regarding a multi-faceted approach to recombinant protein production processes by P. pastoris; including gaining improved understanding of the metabolic pathways involved, accounting for variations in transcriptional and translational efficiency at the single cell level and efficient monitoring and control of methanol levels at the bioreactor level.

Intensified design of experiments for upstream bioreactors

Statistical Design of Experiments (DoE) is a widely adopted methodology in upstream bioprocess development (and generally across industries) to obtain experimental data from which the impact of independent variables (factors) on the process response can be inferred. In this work, a method is proposed that reduces the total number of experiments suggested by a traditional DoE. The method allows the evaluation of several DoE combinations to be compressed into a reduced number of experiments, which is referred to as intensified Design of Experiments (iDoE). In this paper, the iDoE is used to develop a dynamic hybrid model (consisting of differential equations and a feedforward artificial neural network) for data generated from a simulated Escherichia coli fermentation. For the case study presented, the results suggest that the total number of experiments could be reduced by about 40% when compared to traditional DoE. An additional benefit is the simultaneous development of an appropriate dynamic model which can be used in both, process optimization and control studies.

Hybrid metabolic flux analysis and recombinant protein prediction in Pichia pastoris X-33 cultures expressing a single-chain antibody fragment

Despite the growing importance of the Pichia pastoris expression system as industrial workhorse, the literature is almost absent in systematic studies on how culture medium composition affects central carbon fluxes and heterologous protein expression. In this study we investigate how 26 variations of the BSM+PTM1 medium impact central carbon fluxes and protein expression in a P. pastoris X-33 strain expressing a single-chain antibody fragment. To achieve this goal, we adopted a hybrid metabolic flux analysis (MFA) methodology, which is a modification of standard MFA to predict the rate of synthesis of recombinant proteins. Hybrid MFA combines the traditional parametric estimation of central carbon fluxes with non-parametric statistical modeling of product-related quantitative or qualitative measurements as a function of central carbon fluxes. It was observed that protein yield variability was 53.6 % (relative standard deviation) among the different experiments. Protein yield is much more sensitive to medium composition than biomass growth, which is mainly determined by the carbon source availability and main salts. Hybrid MFA was able to describe accurately the protein yield with normalized RMSE of 6.3 % over 5 independent experiments. The metabolic state that promotes high protein yields is characterized by high overall metabolic rates through main central carbon pathways concomitantly with a relative shift of carbon flux from biosynthetic towards energy generating pathways.

Harnessing QbD, Programming Languages, and Automation for Reproducible Biology

Building robust manufacturing processes from biological components is a task that is highly complex and requires sophisticated tools to describe processes, inputs, and measurements and administrate management of knowledge, data, and materials. We argue that for bioengineering to fully access biological potential, it will require application of statistically designed experiments to derive detailed empirical models of underlying systems. This requires execution of large-scale structured experimentation for which laboratory automation is necessary. This requires development of expressive, high-level languages that allow reusability of protocols, characterization of their reliability, and a change in focus from implementation details to functional properties. We review recent developments in these areas and identify what we believe is an exciting trend that promises to revolutionize biotechnology.

Analysis of culture media screening data by projection to latent pathways: The case of Pichia pastoris X-33

Cell culture media formulations contain hundreds of individual components in water solutions which have complex interactions with metabolic pathways. The currently used statistical design methods are empirical and very limited to explore such a large design space. In a previous work we developed a computational method called projection to latent pathways (PLP), which was conceived to maximize covariance between envirome and fluxome data under the constraint of metabolic network elementary flux modes (EFM). More specifically, PLP identifies a minimal set of EFMs (i.e., pathways) with the highest possible correlation with envirome and fluxome measurements. In this paper we extend the concept for the analysis of culture media screening data to investigate how culture medium components up-regulate or down-regulate key metabolic pathways. A Pichia pastoris X-33 strain was cultivated in 26 shake flask experiments with variations in trace elements concentrations and basal medium dilution, based on the standard BSM+PTM1 medium. PLP identified 3 EFMs (growth, maintenance and by-product formation) describing 98.8% of the variance in observed fluxes. Furthermore, PLP presented an overall predictive power comparable to that of PLS regression. Our results show iron and manganese at concentrations close to the PTM1 standard inhibit overall metabolic activity, while the main salts concentration (BSM) affected mainly energy expenditures for cellular maintenance.