I was interviewed by the news magazine, Genetic Engineering & Biotechnology News, on challenges and opportunities of developing and using AI in bioprocessing; see the article entitled ¡°Tackling AI Bottlenecks in Bioprocessing" published on October 9, 2024. Click here for the full article.
- (1) Purely data-driven AI/ML models, without incorporating mechanism insights, lack the capability on interpretability, data fusion, biomanufacturing systems integration, and sample efficient learning.
- (2) Emerging advanced sensing technologies, such as optical sensors and multi-omics assays, can facilitate real-time monitoring of bioprocess dynamics at single molecule and cell scale (such as intracellular metabolites and cell morphology).
- (3) Multi-scale hybrid (mechanistic + statistical) model, characterizing causal dependencies from molecular to cellular to macroscopic scale, can decode the fundamental mechanisms of bioprocessing and support data integration.
- (4) Optimal learning strategies, balancing exploration and exploitation through correctly accounting for all sources of uncertainties, can guide the most-informative design of experiments and support process optimal robust control.
Since induced pluripotent stem cells (iPSCs) have the potential to differentiate into any cell type in the body, large-scale manufacturing of iPSCs is essential for cell therapies and regenerative medicines. I talk about multi-scale bioprocess modeling and risk-based PATs for iPSC cultures in an article at Genetic Engineering and Biotechnology News, which is published on December 6, 2023. Click here for the full article.
- (1) Multi-scale mechanistic model for iPSC cultures can facilitate data integration, including heterogeneous data collected from different production processes with different scales, dynamics, and feeding strategies.
- (2) Unlike CHO cells for mAbs manufacturing, iPSCs cultured in aggregates have heterogeneous micro-environmental conditions shaped by cell-to-cell interaction and they tend to develop different functional behaviors. Thus, production process optimization and cell product quality control require a better understanding on the underlying mechanisms and uncertainties of end-to-end culture process that can facilitate large-scale manufacturing and real-time release.
- (3) In 2D monolayer cultures conducted in the lab, cells attach to the internal surface of petri-dish, which allows us to collect multi-omics and phenotype data and predict singe cell response to environmental changes. This can serve as the building block for assembling different manufacturing processes, including iPSC aggregate cultures.
- (4) This modeling philosophy and process analytical technology (PAT) is extendable and applicable to general biological ecosystems, accounting for complex interaction mechanisms and inherent stochasticity.
RNA structure and functional dynamics play fundamental roles in controlling biological systems and accelerate drug discovery and manufacturing. I talk about a hybrid approach combining mechanistic modeling and ML statistical methods in an article at Genetic Engineering and Biotechnology News,which is published on June 21, 2023. Click here for the full article.
- (Q1:) Why did you use a hybrid (mechanistic + machine learning) model?
- (A1:) This hybrid modeling strategy can extract the key mechanistic structure information of biological systems to reduce model complexity, while keeping scientific interpretability and prediction power. It allows for a more feasible model; but supports a deeper understanding on biomolecular structure-function dynamics, interactions, and enzymatic reaction network mechanisms across multiple scales.
- (Q2:) What is the most interesting and new thing about RNA that this model revealed?
- (A2:) The proposed hybrid modeling strategy can provide insight into energetics and dynamics of enzymatic reactions and RNA conformational change. It can integrate various gene and protein expression measures collected at molecular and macroscopic levels and advance the knowledge of RNA manufacturing mechanisms and guide simultaneous design/control strategies at different levels.
- (Q3:) How could a commercial bioprocessor make use of this information and what would be the key benefits?
- (A3:) First, the hybrid model can accelerate new drug discovery, manufacturing and quick response to virus mutations. Second, it allows us to leverage the information to develop digital twins for enzymatic reaction networks and accelerate design and control of production process.
In biopharmaceutical manufacturing the interactions between cells, nutrients, and reagents in culture determine product quality. The big challenge for process developers is modeling these complex relationships while allowing for variability in materials and other inputs. I talk about bioprocess uncertainty in an article at Genetic Engineering and Biotechnology News,which is published on March 8, 2023. Click here for the full article.
- (Q1:) What is uncertainty analysis and why is it necessary during process development?
- (A1:) Uncertainty analysis, studying the output variability due to the variability of inputs, can support end-to-end decision-making and facilitate bioprocess innovations. It plays a fundamental role to support the FDA¡¯s requirements on Quality-by-Design (QbD) and more.
- (Q2:) What is the approach you and your colleagues propose and how does it differ from current methods? What are the potential benefits of your approach for biopharmaceutical manufacturers?
- (A2:) We introduce a bioprocess knowledge graph (KG) hybrid model and risk analysis framework. Our approach can overcome the limits of current approaches and improve industry practices. It leads to identification of critical input factors and guidance on process monitoring and QC/QA testing.
- (3Q:) How easily could a biopharmaceutical company use your strategy? What expertise, data and data gather technologies are required?
- (A3:) We develop process analytical technology (PAT) software and online training platform with a user-friendly interface to support workforce learning from associate level to expert level. It includes both people, process and tool modules.
With increasingly complex therapies, bioprocessors face a challenge in avoiding errors. I talk about ways of avoiding errors by thoroughly understanding bioprocess mechanisms in an article at Genetic Engineering and Biotechnology News,which is published on November 8, 2022. Click here for the full article.
- (Q1:) Currently, what are the main consequences of lacking a good understanding of the bioprocessing mechanisms behind a drug?
- (A1:) The consequences are lack of integration in both biomanufacturing processes and decision-making and lack of flexibility and automation in biomanufacturing process. Moreover, the time to the market for the final products is lengthy.
- (Q2:) Briefly, how could a bioprocessor better represent the bioprocessing mechanisms behind a drug?
- (A2:) The knowledge graph hybrid model, as a knowledge representation of end-to-end biomanufacturing process mechanisms, can serve as ¡°brain of bioprocessor¡±. Together with advanced sensor monitoring and assay technologies, they allow us to advance the understanding of complex interactions at various levels, support better decision-making.
- (Q3:) What would it take for a bioprocessor to apply this improved representation to the commercial bioprocessing of a drug?
- (A3:)The knowledge graph based digital twin system including advanced process analytical technologies (PATs) can be used by bioprocessors. The calibrated digital twin highly simulates the real biomanufacturing processes and can support drug discovery, root cause analysis, predictive analysis, and bioprocess automation.
Northeastern Global News article, entitled ¡°Northeastern scientists propose AI framework for mass-manufacturing of stem cells for regenerative medicine," reported our novel AI framework for mass-manufacturing of regenerative medicines and cell therapies, which is published on April 3, 2024. Click here for the full article.
- (1) Considering each cell is a complex system composed of metabolic and gene networks, our proposed Biological System-of-Systems (Bio-SoS) framework for Induced Pluripotent Stem Cell (iPSC) cultures can improve the understanding and the prediction of fundamental biological/physical/chemical mechanisms happening in a single cell and in a complex ecosystem composed of cluster of cells. This can guide us to achieve successful cultivation of healthy human induced pluripotent stem cells with consistent quality.
- (2) The developed Bio-SoS is a multi-scale mechanistic model characterizing causal interdependencies at individual cell, aggregate, and cell population levels. It can facilitate data and information integration of various distributed biomanufacturing processes.
- (3) To support large-scale manufacturing, the scientists use suspension bioreactors to culture iPSCs in aggregates. This study could potentially throw light on how embryonic stem cells grow and differentiate during the embryo development and how adult stem cells replace damaged cells to improve public health and speed up new drug development.