Design of experiment (DOE) applied to artificial neural network architecture enables rapid bioprocess improvement

Antibody glycan quality predicted from CHO cell culture media markers and machine learning

N-glycosylation can have a profound effect on the quality of mAb therapeutics. In biomanufacturing, one of the ways to influence N-glycosylation patterns is by altering the media used to grow mAb cell expression systems. Here, we explore the potential of machine learning (ML) to forecast the abundances of N-glycan types based on variables related to the growth media. The ML models exploit a dataset consisting of detailed glycomic characterisation of Anti-HER fed-batch bioreactor cell cultures measured daily under 12 different culture conditions, such as changes in levels of dissolved oxygen, pH, temperature, and the use of two different commercially available media. By performing spent media quantitation and subsequent calculation of pseudo cell consumption rates (termed media markers) as inputs to the ML model, we were able to demonstrate a small subset of media markers (18 selected out of 167 mass spectrometry peaks) in a Chinese Hamster Ovary (CHO) cell cultures are important to model N-glycan relative abundances (Regression - correlations between 0.80-0.92; Classification - AUC between 75.0-97.2). The performances suggest the ML models can infer N-glycan critical quality attributes from extracellular media as a proxy. Given its accuracy, we envisage its potential applications in biomaufactucuring, especially in areas of process development, downstream and upstream bioprocessing.

Machine Learning-Enabled NIR Spectroscopy. Part 3: Hyperparameter by Design (HyD) Based ANN-MLP Optimization, Model Generalizability, and Model Transferability

Data variations, library changes, and poorly tuned hyperparameters can cause failures in data-driven modelling. In such scenarios, model drift, a gradual shift in model performance, can lead to inaccurate predictions. Monitoring and mitigating drift are vital to maintain model effectiveness. USFDA and ICH regulate pharmaceutical variation with scientific risk-based approaches. In this study, the hyperparameter optimization for the Artificial Neural Network Multilayer Perceptron (ANN-MLP) was investigated using open-source data. The design of experiments (DoE) approach in combination with target drift prediction and statistical process control (SPC) was employed to achieve this objective. First, pre-screening and optimization DoEs were conducted on lab-scale data, serving as internal validation data, to identify the design space and control space. The regression performance metrics were carefully monitored to ensure the right set of hyperparameters was selected, optimizing the modelling time and storage requirements. Before extending the analysis to external validation data, a drift analysis on the target variable was performed. This aimed to determine if the external data fell within the studied range or required retraining of the model. Although a drift was observed, the external data remained well within the range of the internal validation data. Subsequently, trend analysis and process monitoring for the mean absolute error of the active content were conducted. The combined use of DoE, drift analysis, and SPC enabled trend analysis, ensuring that both current and external validation data met acceptance criteria. Out-of-specification and process control limits were determined, providing valuable insights into the model's performance and overall reliability. This comprehensive approach allowed for robust hyperparameter optimization and effective management of model lifecycle, crucial in achieving accurate and dependable predictions in various real-world applications.

Data-driven prediction models for forecasting multistep ahead profiles of mammalian cell culture toward bioprocess digital twins

Recently, the advancement in process analytical technology and artificial intelligence (AI) has enabled the generation of enormous culture data sets from biomanufacturing processes that produce various recombinant therapeutic proteins (RTPs), such as monoclonal antibodies (mAbs). Thus, now it is very important to exploit them for the enhanced reliability, efficiency, and consistency of the RTP-producing culture processes and for the reduced incipient or abrupt faults. It is achievable by AI-based data-driven models (DDMs), which allow us to correlate biological and process conditions and cell culture states. In this work, we provide practical guidelines for choosing the best combination of model elements to design and implement successful DDMs for given hypothetical in-line data sets during mAb-producing Chinese hamster ovary cell culture, as such enabling us to forecast dynamic behaviors of culture performance such as viable cell density, mAb titer as well as glucose, lactate and ammonia concentrations. To do so, we created DDMs that balance computational load with model accuracy and reliability by identifying the best combination of multistep ahead forecasting strategies, input features, and AI algorithms, which is potentially applicable to implementation of interactive DDM within bioprocess digital twins. We believe this systematic study can help bioprocess engineers start developing predictive DDMs with their own data sets and learn how their cell cultures behave in near future, thereby rendering proactive decision possible.

Machine learning in bioprocess development: from promise to practice

Fostered by novel analytical techniques, digitalization, and automation, modern bioprocess development provides large amounts of heterogeneous experimental data, containing valuable process information. In this context, data-driven methods like machine learning (ML) approaches have great potential to rationally explore large design spaces while exploiting experimental facilities most efficiently. Herein we demonstrate how ML methods have been applied so far in bioprocess development, especially in strain engineering and selection, bioprocess optimization, scale-up, monitoring, and control of bioprocesses. For each topic, we will highlight successful application cases, current challenges, and point out domains that can potentially benefit from technology transfer and further progress in the field of ML.

Comparative study of lipid nanoparticle-based mRNA vaccine bioprocess with machine learning and combinatorial artificial neural network-design of experiment approach

To develop a combinatorial artificial-neural-network design-of-experiment (ANN-DOE) model, the effect of ionizable lipid, an ionizable lipid-to-cholesterol ratio, N/P ratio, flow rate ratio (FRR), and total flow rate (TFR) on the outcome responses of mRNA-LNP vaccine were evaluated using a definitive screening design (DSD) and machine learning (ML) algorithms. Particle size (PS), PDI, zeta potential (ZP), and encapsulation efficiency (EE) of mRNA-LNP were optimized within a defined constraint (PS 40-100 nm, PDI≤0.30, ZP≥(±)0.30 mV, EE≥70%), fed to ML algorithms (XGBoost, bootstrap forest, support vector machines, k-nearest neighbors, generalized regression-Lasso, ANN) and prediction was compared to ANN-DOE model. Increased FRR decreased the PS and increased ZP, while increased TFR increased PDI and ZP. Similarly, DOTAP and DOTMA produced higher ZP and EE. Particularly, a cationic ionizable lipid with an N/P ratio ≥6 provided a higher EE. ANN showed better predictive ability (R²=0.7269-0.9946), while XGBoost demonstrated better RASE (0.2833-2.9817). The ANN-DOE model outperformed both optimized ML models by R²=1.21% and RASE=43.51% (PS prediction), R²=0.23% and RASE=3.47% (PDI prediction), R²=5.73% and RASE=27.95% (ZP prediction), and R²=0.87% and RASE=36.95% (EE prediction), respectively, which demonstrated that ANN-DOE model was superior in predicting the bioprocess compared to independent models.

On-line monitoring of total sugar during kombucha fermentation process by near-infrared spectroscopy: Comparison of linear and non-linear multiple calibration methods

Kombucha is widely recognized for its health benefits, and it facilitates high-quality transformation and utilization of tea during the fermentation process. Implementing on-line monitoring for the kombucha production process is crucial to promote the valuable utilization of low-quality tea residue. Near-infrared (NIR) spectroscopy, together with partial least squares (PLS), backpropagation neural network (BPANN), and their combination (PLS-BPANN), were utilized in this study to monitor the total sugar of kombucha. In all, 16 mathematical models were constructed and assessed. The results demonstrate that the PLS-BPANN model is superior to all others, with a determination coefficient (R²p) of 0.9437 and a root mean square error of prediction (RMSEP) of 0.8600 g/L and a good verification effect. The results suggest that NIR coupled with PLS-BPANN can be used as a non-destructive and on-line technique to monitor total sugar changes.

Harnessing the potential of machine learning for advancing "Quality by Design" in biomanufacturing

Ensuring consistent high yields and product quality are key challenges in biomanufacturing. Even minor deviations in critical process parameters (CPPs) such as media and feed compositions can significantly affect product critical quality attributes (CQAs). To identify CPPs and their interdependencies with product yield and CQAs, design of experiments, and multivariate statistical approaches are typically used in industry. Although these models can predict the effect of CPPs on product yield, there is room to improve CQA prediction performance by capturing the complex relationships in high-dimensional data. In this regard, machine learning (ML) approaches offer immense potential in handling non-linear datasets and thus are able to identify new CPPs that could effectively predict the CQAs. ML techniques can also be synergized with mechanistic models as a ‘hybrid ML’ or ‘white box ML’ to identify how CPPs affect the product yield and quality mechanistically, thus enabling rational design and control of the bioprocess. In this review, we describe the role of statistical modeling in Quality by Design (QbD) for biomanufacturing, and provide a generic outline on how relevant ML can be used to meaningfully analyze bioprocessing datasets. We then offer our perspectives on how relevant use of ML can accelerate the implementation of systematic QbD within the biopharma 4.0 paradigm.

Clever Experimental Designs: Shortcuts for Better iPSC Differentiation

For practical use of pluripotent stem cells (PSCs) for disease modelling, drug screening, and regenerative medicine, the cell differentiation process needs to be properly refined to generate end products with consistent and high quality. To construct and optimize a robust cell-induction process, a myriad of cell culture conditions should be considered. In contrast to inefficient brute-force screening, statistical design of experiments (DOE) approaches, such as factorial design, orthogonal array design, response surface methodology (RSM), definitive screening design (DSD), and mixture design, enable efficient and strategic screening of conditions in smaller experimental runs through multifactorial screening and/or quantitative modeling. Although DOE has become routinely utilized in the bioengineering and pharmaceutical fields, the imminent need of more detailed cell-lineage specification, complex organoid construction, and a stable supply of qualified cell-derived material requires expedition of DOE utilization in stem cell bioprocessing. This review summarizes DOE-based cell culture optimizations of PSCs, mesenchymal stem cells (MSCs), hematopoietic stem cells (HSCs), and Chinese hamster ovary (CHO) cells, which guide effective research and development of PSC-derived materials for academic and industrial applications.