In bioprocess development, could you reduce the need for 60 to 80% of experiments while achieving more accurate results? The answer might lie in a revolutionary approach called hybrid modeling, which combines mechanistic knowledge with cutting-edge machine learning to accelerate the development of biopharmaceutical processes.
In this article, we explore the insights shared by Michael Sokolov, Co-founder and COO of DataHow, on the transformative power of hybrid modeling in bioprocessing. This concept is discussed in greater detail in an episode of the Smart Biotech Scientist Podcast, hosted by David Brühlmann, founder of Brühlmann Consulting.
The Power Of Digital Technology In Bioprocess Development
If you're working in bioprocessing, the future is clear: digital technology should be an integral part of your process development strategy. According to Michael Sokolov, embracing digital tools to analyze data is not just beneficial—it’s essential for unlocking faster and more insightful results.
DataHow, a spin-off company from ETH Zurich, specializes in advanced process data analytics and modeling, particularly in the biopharmaceutical and chemical industries. Michael Sokolov has witnessed firsthand how leveraging hybrid models can not only speed up the development cycle but also reduce the reliance on costly and time-consuming experiments.
You should trust digital technology more and use it as an integral part of your daily work. Eventually, it should allow you to conduct fewer experiments and still derive meaningful insights.
The Key Challenges In Bioprocessing
Before diving into the benefits of hybrid modeling, it’s important to understand the core challenges faced by biotech professionals in the bioprocess development space.
1. Time And Cost Pressures
Developing a bioprocess is a race against time and budget. Companies need to optimize the process as quickly and cost-effectively as possible, selecting the right cell lines, reactor types, and operational parameters to scale up toward manufacturing. Traditional methods are often slow, requiring many experimental iterations before finding the optimal solution.
2. Minimizing Failure Rates
Once a bioprocess is developed, ensuring its robustness during manufacturing is critical. Biopharmaceutical companies face an average 1% failure rate in manufacturing, which can lead to substantial financial losses. Reducing failure rates and shortening the time-to-market are therefore vital objectives, especially for smaller biotech companies looking to introduce novel therapies quickly and cost-effectively.
Hybrid Models: The Future Of Bioprocess Development
The concept of hybrid modeling offers a powerful solution to these challenges. Hybrid models combine mechanistic knowledge—the engineering and biological principles that govern bioprocesses—with machine learning (ML), allowing for more precise modeling even with incomplete data.
Hybrid Modeling Explained
Hybrid models work on the premise that we have two key insights:
- We know our bioprocess to some extent, but we don’t have complete knowledge.
- We can improve our process understanding by leveraging both mechanistic and data-driven insights.
The idea is to combine incomplete data with mechanistic insights to answer practical questions about the bioprocess. By using machine learning to fine-tune the model, hybrid models help scientists make more informed decisions, speeding up the development process with fewer experiments.
The idea of hybrid modeling is therefore to say, well, we have an imperfect knowledge or an incomplete knowledge and also an incomplete data set available, but these two combined should be leveraged to the maximal degree.
Hybrid Modeling vs. Traditional Methods
Traditional modeling methods, such as linear regression, have long been the go-to for bioprocessing. However, these models struggle to capture the nonlinear relationships that define modern bioprocesses. Biological systems are highly complex and dynamic, and the interrelationship between design parameters (e.g., temperature, pH, cell line) and product quality attributes is rarely linear.
Traditional Models: Limited By Simple Tools And Linear Methods
In a traditional setup, scientists often conduct numerous experiments to identify these relationships. However, this approach is resource-intensive and time-consuming, especially when data is scarce.
Hybrid Models: Nonlinear And Adaptable
Hybrid models address these challenges by introducing nonlinear mechanistic models that act as a backbone. These models are then enhanced by machine learning tools that optimize them for specific bioprocesses. This combination of approaches allows scientists to understand processes with much less data, requiring fewer experiments to achieve reliable insights.
Benefits Of Hybrid Modeling In Bioprocess Development
The primary advantages of hybrid modeling include:
- Reduced Data Requirements: Hybrid models enable deeper insights even with limited data, which can dramatically reduce the number of required experiments.
- Faster Process Development: Using data-driven decisions to guide experimental design, hybrid models accelerate the development cycle.
- Increased Understanding: Machine learning helps scientists uncover hidden relationships and optimize processes more effectively.
In essence, hybrid modeling facilitates complex decision-making while requiring less data. It’s a tool that helps scientists make better decisions in real time, ultimately improving the understanding and efficiency of bioprocesses.
From Gut-Based Decisions To Data-Driven Insights
Historically, many decisions in bioprocess development have been based on the experience and intuition of scientists—what Michael Sokolov refers to as “gut-based decisions.”
While experience is valuable, it’s increasingly important to replace these subjective judgments with data-driven decisions. With hybrid models, scientists can leverage machine learning to make more educated, well-qualified decisions based on data rather than gut feelings.
The Impact: Reducing Experiments By 60-80%
DataHow’s work with large pharmaceutical companies and contract development and manufacturing organizations (CDMOs) has demonstrated that hybrid modeling can reduce the need for experiments by 60 to 80%. This is especially valuable in the early stages of process development, from initial design of experiments to technology transfer and manufacturing.
While this reduction in experimental work leads to significant savings in both time and costs, it’s instrumental in emerging therapies, where the status quo doesn’t apply, and each batch can be much more expensive to produce.
A Small Biotech Use Case
Consider a small biotech startup with limited resources and a small reactor capacity. They begin by running a design of experiments (DoE) to explore the process parameters (e.g., pH, temperature, cell line). As they generate data, a hybrid model helps them analyze it and suggest the next best steps, enabling them to focus resources on the most promising experiments. Over time, this iterative, data-driven approach leads them to an optimal process design that meets their specific goals.
The Digital Twin: Bridging The Gap Between Theory And Practice
A powerful extension of hybrid modeling is the concept of the digital twin. Essentially, a digital twin is a real-time simulation of the bioprocess, continuously learning and adjusting based on incoming data from the process itself.
In real-time, the digital twin can:
- Fetch real-time data from reactors or IT systems.
- Process this data and make decisions about how to optimize the process.
- Feed this information back into the system, enabling adaptive control and reducing manual intervention.
This closed-loop system helps maintain optimal process performance and offers valuable insights into how the process can be fine-tuned, even as it’s running.
The idea of a digital twin is that only have a model which you trust in understanding your process and withdrawing relevant information of how to design it. But that sparing partner would be active in real time. That means you connect it in two directions. First of all, it fetches the data real time from your IT environment or from the reactors directly it processes them. It withdraws an important decision of how to operate the process. And instead of only displaying it on the screen, it feeds it back directly as a control loop into the system, allowing you to have less manual operations and a more adaptive nature. As in real time. You're actually able to to always link the running system towards the learning model and back towards an optimally controlled system.
The Challenges And Limitations Of Hybrid Modeling
While hybrid modeling offers many benefits, there are a few key considerations:
- Goal Definition: The effectiveness of hybrid modeling relies on clearly defining the goals upfront. Without a specific and measurable goal, the model won’t provide useful insights.
- Change Management: Hybrid models can’t solve everything. It’s essential to focus on the factors that have the greatest impact on process performance and avoid unnecessary changes.
- Learning Curve: Hybrid modeling tools require significant training to be effective. Scientists must invest time in mastering these tools for them to become valuable assets.
Regulatory Considerations
As the Industry 4.0 movement grows, regulatory bodies are open to new technologies like digital twins and hybrid modeling. While the GMP validation is required for active process control in manufacturing, using these tools for monitoring and process design doesn’t necessarily require full regulatory approval.
Conclusion
In a world where time and resources are limited, hybrid modeling offers a smarter, more efficient way to develop and scale bioprocesses. By combining mechanistic knowledge with machine learning, it allows biopharma companies to reduce the number of experiments, optimize processes faster, and gain deeper insights into their operations.
As more biotech firms adopt hybrid models and digital twin technology, we’ll see accelerated timelines, reduced costs, and the ability to bring more life-saving therapies to market. The future of bioprocess development is digital, and hybrid modeling is at the forefront of this transformation.
About Michael Sokolov
Michael Sokolov is the s co-founder and COO of DataHow AG, a spin-off company from ETH Zurich specializing in process data analytics and modeling with a particular focus on the biopharmaceutical and chemical domains. His main activities are centered on managing data analytics and DataHowLab software projects with global pharma accounts and coordinating all operational and financial activities of DataHow.
Connect with Michael Sokolov on LinkedIn.
David Brühlmann is a strategic advisor who helps C-level biotech leaders reduce development and manufacturing costs to make life-saving therapies accessible to more patients worldwide.
He is also a biotech technology innovation coach, technology transfer leader, and host of the Smart Biotech Scientist podcast—the go-to podcast for biotech scientists who want to master biopharma CMC development and biomanufacturing.
Hear It From The Horse’s Mouth
Want to listen to the full interview? Go to Smart Biotech Scientist Podcast.
Want to hear more? Do visit the podcast page and check out other episodes.
Do you wish to simplify your biologics drug development project? Contact Us
