BioPharm International: Hi, I'm Amber Lowry, Senior Editor of Special Projects for BioPharm International, and I'm here with Gunner Malmquist, Senior Principal Scientist, and Nick Whitelock, Sales Specialist, both at Cytiva. Thank you both for being here today.
Nick Whitelock (Cytiva): Thank you for having us.
Gunner Malmquist (Cytiva): Thank you, Amber.
BioPharm International: Great. So, today we're going to discuss insider recommendations for getting started with mechanistic modeling of chromatography. So, let's dive in. Can you please quickly recap, in a few words, what mechanistic modeling of chromatography actually is?
Malmquist: Yes. If we think of this as being a way of simulating chromatograms with the help of the computer, you take data related to your chromatography system, to your column, and to the separation that you're doing and put that into the computer. And there, algorithms and equations are used to describe the phenomena happening. So you use differential equations to describe mass transfer and adsorption isotherms to describe the binding event. Out of the computer comes simulated chromatograms—1,000 chromatograms in a few minutes.
And the good news is that you are able to do this, not only for your molecule of interest, but also for all the impurities.
Nick, how do you do this in reality?
Whitelock: Absolutely. So primarily, we separate the emergent complex behavior of chromatography into the rather simple, fundamental physical–chemical phenomena involved. So of course, fluid dynamics is crucial here—both the fluid dynamics within the chromatography system you are using and, of course, the column. We can counterize those with very simple pulse experiments. We're talking something akin to an HETP test, with and without a column, and potentially with a large non-pore-penetrating tracer as well. That completely captures all of the fluid dynamics and, of course, the internal geometry of the column and system.
Once we've defined these fluid mechanical effects, we can then start to provide an insight into binding thermodynamics and kinetics. This we can do with rather simple, gradient elution experiments for thermodynamics and supplement this with perhaps isocratic elutions to fit any kinetic resistances to transport, saturation behaviors with any overloaded runs, and of course invoke a pH dependency later if required. We take all of this data, we leave the model to fit the real-world physical parameters, so diffusion rates and porosities with an inverse method. So that simply, we turn out thousands of runs and find that set of parameters that best describes it.
Finally, we would recommend doing a verification of your model because, of course, without demonstrating it is predictive and captures the dynamics, it's not much use. So we do this by looking at the visual quality of fit, by looking at the confidence intervals of these fitted parameters, and of course propagating them through to any model predictions. And finally, of course, the gold standard of any modeling project is that experimental verification.
After this, of course, we have built a profound digital twin of your process, at which point we can run experiments—many thousands of them—in silico in a few seconds.
BioPharm International: So what would you both say is the greatest obstacle to implementing mechanistic modeling?
Whitelock: I would say lack of established mechanistic modeling expertise in the industry. Of course, we're dealing with quite a modern endeavor. In a sense, mechanistic modeling itself is quite old, but mechanistic modeling of biopharmaceuticals has come came to prominence in recent years. And frankly for a lot of companies, there isn't that established skill base out there. Of course, to assist in that, so Cytiva does provide extensive training, tutorial files, help texts, and, of course, user guides and consultancy services to help meet this skills gap.
Malmquist: I think another thing that comes into my mind is that for someone who has been doing process development the traditional way using DoE, this is a new workflow. You need to do different experiments, and that could take some time to learn. They are easy to learn, but it's hard to change your usual ways of working. But it's also the reliance on analytical colleagues to assay your fractions, and you need to establish a good relationship with your analytical colleagues to get successful in this.
BioPharm International: Excellent. And given those challenges, what would you say are the top practical tips that you would give someone who's planning to implement mechanistic modeling into their process development (PD) work? So, dos and the don'ts.
Whitelock: I would say focus on data quality rather than data quantity. So it's rather important to have the best means of generating this understanding of the system dynamics. Of course, Cytiva does also provide tools to help in that regard, such as precalibrated columns. But unlike a statistical model, really it's having that singular correct measure of the systems is the crucial part, rather than a lot of data to average out the noise.
Malmquist: Another tip is to make sure that you account for the need to validate your model. Once you have calibrated it, you have your equations all in place, you need to do that crucial experiment that proves that your model actually is able to predict reality. That needs to be factored in.
Another very good [piece of] advice is to reach out. There are advice and expertise available that can help you on this journey, and Cytiva is happy to help out.
Whitelock: Of course, we have an ever-growing community of modeling experts, which has been grown by Cytiva, an ever-growing library of papers, ever more case studies, help text, publications, and so forth. So the barrier to entry is always reducing and the community out there is ever-growing.
BioPharm International: Those are great points. And finally, what would you both say is the biggest benefit of in silico process development driven by mechanistic models?
Malmquist: My main benefit or driver to go here would be the increased process understanding. Having a model that describes your unit operation allows you to explore edges of failure. It allows you to identify critical process parameters in a much better way than any DoE model would ever do.
Whitelock: And to add to that, I would also say efficiency as well. So the fact we can generate this process understanding with an incredibly lean experimental burden. I believe Boehringer Ingelheim recently published that they experienced a 75% reduction in the process characterization activities using mechanistic models versus a conventional design of experiments approach. Additionally, in terms of material use as well, which of course translates from reduced experimental burden, but ultimately we can find a mechanistic optimum in terms of process performance and of course extrapolate out to unseen process conditions.
So really it's a combination of efficiency, both in terms of process development and ultimately process efficiency as well.
BioPharm International: Great. Well, unfortunately that's all the time we have for today. But again, thank you both so much for joining us today and for sharing your insights with us.
Whitelock: Thank you.
Malmquist: Thank you.