Fractals:

Click Here for podcast episode details and listening links.

 

[Colin Miller] 

Hello, I'm Colin Miller, CEO at the Bracken Group and this is Fractals, Life Science Conversations. Bracken is the professional services firm for life sciences and digital health organizations. Our intelligence ecosystem fulfils consulting, regulatory, marketing and analytics needs with an integrated and strategic approach.

Today's episode will focus on the science and engineering behind the pharmaceutical coating process and how advanced modelling and simulation are transforming the ways we design, optimize and scale tablet manufacturing. To explore this topic, I'm delighted to be joined by Dr. Michael Choi, founder and president of Particle Coating Technology Solutions, or PCTS. Michael lives in the same geographical area as I do and we've met multiple times at networking events and I'm delighted to have him join me today on this podcast.

Michael is a process expert and the inventor of industry-leading simulation tools used to model and optimize tablet coating. His doctoral research focused on the modelling and experimental verification of both batch and continuous coating processes, laying the groundwork for a career dedicated to improving the scientific rigor and predictability of pharmaceutical manufacturing. Michael also holds a patent on tablet coating uniformity simulation, a tool that has helped engineers better understand and control one of the most complex steps in solid oral dosage production.

Michael, welcome to Fractals and thank you for joining me today.

 

[Michael Choi]

Colin, thanks so much for inviting me. It's a pleasure and honor to be here.

 

[Colin Miller]

No, it's a pleasure to have you here and we've talked multiple times at networking events locally, so from my perspective, this is delightful, we actually have an in-depth conversation. And you've held multiple leadership roles across technical operations, supply chain and manufacturing in pharmaceutical companies. Have these experiences influenced your perspective on the challenges of process modelling and optimization?

 

[Michael Choi]

I guess when I started my career, I was very technical. I was very engineering centric. So my focus in modelling was almost entirely on scientific integrity of the model.

So that would include things like accuracy, precision and mechanistic correctness. And I cared deeply about whether the model truly reflected the physics and chemistry of the process. But as I moved into broader leadership roles in capital projects, operations, quality and eventually general management, I would say my perspective expanded.

I began to see that model isn't only a scientific tool, it's also a decision making tool. And what I recognized was that different stakeholders were looking for clarity on cost, capacity, quality control and risk, often in very simple and actionable terms. And so I think that experience reshaped how I approach modelling and optimization.

Today, I still insist on scientific rigor, that's not going to change. But I place equal emphasis on how a model informs business decisions, supports scale up and drives operational. So in other words, the optimization component became just as important as the underlying science, because that's where the practical meaning of what I do lies.

 

[Colin Miller]

Fascinating. So thank you. We're going to jump into more of the manufacturing side of it later on.

But there were a couple of comments there that I'm used to hearing in the terms of medical imaging, and in particular, the term accuracy and precision. I'd love to understand how you see it in the context of particle coating. And what are the challenges there?

 

[Michael Choi]

For me, the accuracy and precision, we can use the statistical terms for that. But when we're talking in the context of modelling, accuracy is about the target itself. And are we getting close to the target?

Because modelling will give you an answer. But how accurate it is, how close it is to the target, that is very important. Precision has to do with how repetitively you can get to the same place, and how closely, how tightly that you can get it.

So when I look at it in terms of modelling, I tend to think about, when I talk about precision, the first thing that comes to mind is Monte Carlo simulation, a stochastic approach. And the more you run the simulation, the more, I should say, more precise you're going to get. But there's a cost to that.

The cost is usually computational cost. And it can be real cost, because sometimes we exceed the limits of CPUs that we run. And the area I'm in, there is actually a simulation that involves very heavy computational power that's required to run it.

And that's a new technology. I shouldn't say new, it's been around for a couple of decades now. But it has to do with the discrete element method.

So that's how I look at precision and accuracy.

 

[Colin Miller]

So we used to say, I 100% agree with your definition, I love it, it works perfectly. How do you relate that now to particle coating on tablets?

 

[Michael Choi]

Oh, so coating uniformity has to do with probability. So let me step back and see if I can explain this in layman's terms. The goal of coating process is to have every particle look the same, or at least have the same attributes.

But because of the random nature of any unit operation, you're going to see variation. So coating uniformity requires that every coated tablet are the same within the tolerance, of course. And that tolerance itself, we can look at it as a precision.

The accuracy would be, in this case, would be the assay, or the amount of coating that we actually put on. So if we said, we're going to put three milligrams, then we put three, that would be the accuracy. The precision would be between this tablet and that tablet, what is the variation?

And that to me is precision in coating.

 

[Colin Miller]

I had no idea what goes on. Particle coating is not something I've ever studied. And so I really appreciate that insight.

And of course it makes perfect sense because you can't have tablets all coming with varied level of coating thicknesses and particles and all the rest of it. And so you have to ensure that's all correctly aligned. Now that explanation alone helps me figure out, now you put precision and accuracy together.

And you earlier on mentioned about Monte Carlo simulation. Can you explain a little bit more on that and now how that ties into what is it that you are simulating and how are you building this?

 

[Michael Choi]

So that doesn't tie into the current model. But when I patented the coating process earlier in early 2000s, that was using Monte Carlo method. To give you some context, when I was doing my doctoral thesis, I was doing modeling of the coating process and then doing experiments to verify that.

Well, the modeling itself, for the most part, was fairly straightforward. We used the computational fluid dynamics to actually model the movement of tablets, movement of air, and then we characterized the spray separately. And so it was a matter of putting all these things together to see the interaction and what happens in the interfaces between these three phases.

But then came this time to calculate the coating uniformity. And we're scratching our heads and saying, okay, the model is great. It explains how the temperature changes, how the flow changes from one location to another.

Now how do we put it all together to get a coating uniformity? And my advisor at the time, he was stumped as well. And so it took us a little bit.

And about a couple of weeks later, he said, he came back and says, let's try this Monte Carlo method. And to explain the Monte Carlo method, it's very simple. So suppose we have a wall and there's a circular shape in the wall and light comes through it.

So you can see that it is circle. But suppose now we put in a thick piece of paper across this hole and now we can't see what shape it is. So one way to do it is we pull out our machine gun, which is stored in my shed.

I'm joking, of course. And then we start shooting at the wall. We shoot like 10 bullets and we can see the bullet holes and we can see the light coming through.

We can make out what shape it is behind the paper. But as we shoot more bullets, the shape becomes clearer and clearer. And eventually we shoot a million bullets and we can see that it's a circle.

So that's how I explain Monte Carlo method. It's just a series of random simulation. And then at the end, we put it all together to see what, in this case, what shape it is.

OK, thank you.

 

[Colin Miller]

That's a great definition of Monte Carlo, I have to admit. That was fascinating. Thank you.

And I'll remember to keep a shed and a shotgun or at least a machine gun next time. I don't recommend it. Yeah, no, I agree with you.

So having done a lot of this work and built this simulation out, what inspired you to find the company Particle Coating Technology Solutions? And how has that vision for the company evolved from its beginnings?

 

[Michael Choi]

When I was at Merck, I worked with perhaps amazing people. There were a lot of very smart people and Merck at the time was recruiting best people from best universities. I was one of those guys that kind of slipped through because I had some specialization.

But one thing that kind of frustrated me was that I couldn't get the scientists to use science, even though these guys were extremely bright. And you're looking at the top tier of the people that were out there in the industry. But I couldn't quite convince them to use just simple science that they learned in undergraduate.

So when I left Merck, my motivation for the company was very simple. I wanted to make coating technology more accessible. A division I had was to expand the use of coating in the pharmaceutical industry by grounding it in clear science and practical modeling.

I truly believe that if we could explain the physics in the way the practitioners could actually use, this technology would be adopted. The adoption would grow much faster. So that's why I filed a patent in 2004 to support that mission.

Now, by the time the patent was approved, I was back. I already had stepped into other roles, broader roles across pharmaceutical industry, from technical operations to general management. I guess I gained different perspectives.

It was a much wider view of how technologies would succeed and sometimes fail inside real organizations. So during that period, I watched the coding field evolve from the outside. The science was advancing quickly and the modeling community was making impressive progress.

It was really exciting to see. But then about six years ago, I noticed something of a concern. Despite all the scientific progress, the gap between modeling and practical application was actually widening.

Effective coding simulation now required a small army of specialists. There were coding experts, simulation experts, statisticians all working together. The tools were becoming more powerful, but also more exclusive.

So practitioners, the people who needed these insights the most, were being pushed to the sidelines. And that realization brought me back. I restarted PCTS with the renewed purpose to bridge the gap between advanced modeling and everyday practice.

You could say that my goal was to make coding science intuitive, accessible, and actionable, not just for the scientists, but also for formulators, process engineers, and manufacturing teams that need the answers real time. So in many ways, the mission is the same as 2003. But now it is shaped by a much deeper understanding of how science, operations, and people come together in real manufacturing environments.

 

[Colin Miller]

What a story, what a history and pathway. That's fantastic to use. So I think you've pretty well covered the effective modeling to support easier scale-up from lab production.

I presume the current methodologies must be quite slow compared to everything, because you wouldn't have built this otherwise. So this must save time in the overall process.

 

[Michael Choi]

When I think about effective modeling, I talked about the process science piece, I guess to a great degree, because when I talk about Monte Carlo simulations and mathematical modeling, it talks about the process part. But I also think that effective modeling should also cover operational piece as well. The models that we came up, it doesn't use Monte Carlo method anymore because we found a new way to actually model the process.

And it's a very interesting discovery. This was also published about three, four years ago. And it's about using mathematics to actually replace Monte Carlo method.

So this was, I thought it was a great discovery. It kind of came to the same mathematical series as like exponential functions. So if you look at all the probabilities of particles going into spray zone and amount of coating it gets, it forms certain patterns.

And that pattern can be explained by Taylor series. Now in this particular case, it's a Maclaurin series. But just like the exponential function, you don't know what's the equation behind it, but it's a series.

And we were able to bring that down. So it's very quick to calculate because it uses these functions. Now, what I didn't mention about the effective modeling, I think from process perspective, we want the model to be able to take what we learned from a small scale and drag and drop it directly into commercial scale.

That concept is called one batch scale up. And our digital twin is designed to do something like that. But it does take a little bit of practice, if you will.

It takes a little bit of data and a little bit of training for the program itself. But to me, that is the effective modeling from process side. But from operational piece, I think we also need to look at it from how does it help the operation.

So to give you an example, when we do, say, oral solid dosage manufacturing, we have different unit operations, and the yields vary within each unit operation. So in the case of, say, a film coating process, what happens is that we make the tablets, and then we usually make a large amount of tablets. And we typically have to split that into smaller sub-batches to coat them because it's a big, huge batch.

That means that when we start splitting, what is the batch size we target with varying yields? I think one of the misconceptions a lot of folks have with coating processes, you can simply scale it proportionally. And it doesn't work that way.

You can keep all the conditions the same and coat a shorter period of time, or you can spray slower and coat the same amount of time. Either way, you get a different product. So that's a misconception a lot of people have.

How do we deal with different batch sizes? You use digital twin, and it'll tell you how to get there. So this is one where it can be very helpful.

Okay.

 

[Colin Miller]

Wow. So it's a way more complex area and issue than I'd ever appreciated. Fascinating.

So over the last 10 or 15 years, what's been the most surprising change or development you've seen in the practice of particle coating?

 

[Michael Choi]

In the area of coating, there was a very exciting new technology. I think it's still considered emerging technology, and that's continuous manufacturing. Now, continuous manufacturing doesn't mean that you have to use a continuous process.

It can be batch-wise processed in a continuous way, or it could be semi-batch or semi-continuous, if you will. But when the concept of continuous manufacturing came out, some vendors decided to build these coating technologies that are continuous, fully continuous, meaning that you put tablets on one side, uncoated tablets on one side, and on the other end, we have coated tablets. And this would be done continuously.

They didn't recognize that there is another factor called residence time distribution. So the system is not a plug flow system. Some will, even if you put two tablets at exactly the same time, you cannot guarantee that they come out at the same time.

So there was another factor that they needed to consider. They didn't. And then when they built, spent a ton of money building these new technologies, they realized that there is a quality control limitation.

So that was kind of surprising to me because this is something that can be quickly calculated using models. And we did have to go through all that pain to just find out that there's limited application for that technology. So a lot of time and money could have been saved.

So I guess in some ways, it's a reminder that enthusiasm for new technology sometimes outpaces the fundamentals.

 

[Colin Miller]

Yeah, of course. So where do you see it going in the future? If you have a crystal ball, where's particle coating going?

Where's AI? Is AI involved in this and advanced sensors and time modeling, etc.? How do you see it in the future coming out?

 

[Michael Choi]

I really hope the industry does pick this up. And I'm sure eventually it will get picked up. But how quickly will it get picked up?

So I still think that continuous manufacturing is just a wonderful new technology, right? So even though we're not using continuous process, even if you have to use batches, smaller batches in sequential fashion, it still fits into continuous manufacturing concept. And with the understanding of the coating process, I think we can make significant progress.

One of the concepts that we have with a commercial scale system is that we can produce things faster using bigger equipment. But the model here clearly shows that smaller equipment and smaller batch sizes can deliver a much higher production rate than the large equipment. So that was one of the interesting findings.

And so if we can marry these up, right, digital twin technology and continuous manufacturing, I think the real breakthrough will come with that marriage. So with the digital twin guiding design control and scale up, I think we can really unlock the efficiency and flexibility and even responsiveness for that continuous manufacturing system. Now, AI, right now, I'm using mathematical models.

And in order to do inverse modeling, I have to customize each solution because there are a ton of different solutions that you can do. Well, that you have to go through in inverse modeling. But as we get more specific, when we get more data, I think AI could be useful.

And right now, for inverse modeling, there is institutions that are actually doing a lot of research on it. So when we talk about AI, we're talking about statistical approaches, pattern recognition, so neural networks. We can use those if we have a lot of data, reliable data.

But we're still doing it in a statistical fashion. Doing inverse modeling now gets even more complicated because now we have to reverse it. So there's approaches like Bayesian approaches of inverse modeling.

But again, these require special, I guess, statistical tools. And we're still building those. So when those become available and when there's a lot of data, I think AI could be useful.

There has been a lot of movement towards advanced sensors. One in particular, there's an instrument that came out of research from University of Graz in Austria. That's showing some promise, but there were some limitations as to what type of coatings that can be used.

But those are the things that are really helping with real-time modeling, real-time simulations, and real-time monitoring of the processes. So I think combining all these, I think next decade could bring some step change in how we design and operate coating processes. Fascinating.

 

 [Colin Miller]

Yeah, it's so many advances that are possible. And it's interesting because you've got that insight on the multiple different facets that it's not necessarily one. It's bringing them together.

Wow. Appreciate it. Changing tack slightly, without breaching confidentiality, can you give us a case study where simulation uncovered a problem or solution that wasn't obvious experimentally?

And what impact did that have on the outcome?

 

[Michael Choi]

One example that I'd like to share is the one that when I just joined Merck, there was a troubleshooting situation. We're about to launch a new product, and all we had to do was pass three validation batches. One of those validation batches failed.

Nobody could explain why. I started reviewing the data, and there was a small difference in temperature between the failed batch and the other two batches. And so I quickly did some calculations using a model, thermodynamics model.

And to my surprise, that difference actually explained why we had 10% lower drug assay on this failed batch. Now, it was wonderful that I had this answer, but the next step was even more difficult. Because when I shared this information with an equipment engineer, he initially dismissed it.

So the real situation, he was actually laughing his head off because, hey, what? You saw this difference in temperature, and you're telling me that's going to explain the situation? It took some convincing, but I finally got him to start pulling some data.

So we're looking at the SCADA data, and surely enough, we said this data that shows that the flow meter was drifting. So there was an error with the flow meter, and that was the root cause. And the model had pointed us to the root cause that nobody picked up, because they thought this was well within the range of equipment instrument tolerance.

So the impact here was substantial. It was a matter of canceling the launch, or do we continue with it, right? That was the impact.

So I thought this would be a great example because it uncovers subtle process behaviors that would otherwise go unnoticed. And it shows how powerful modeling can be when paired with solid engineering judgment.

 

 [Colin Miller]

Wow. Wow. So you basically, by applying the modeling, saved a whole production because you found out exactly where the problem was rather than losing a complete set of batches for the FDA.

 

[Michael Choi]

It was a matter of actually losing the product, period. Wow. Because it was a new technology that they were trying, drug layering, and it is very precise technology.

And when work was very scientific, scientifically driven, and when they didn't understand it, they wouldn't go forward with it. And they did have other problems with other products where they couldn't scale up. So this was one that I could contribute immediately.

And it was really the power of modeling. And really, it wasn't that complicated. And if you think about what coding process characterization requires, thermodynamics is only one of the tools, right?

One of the sciences. There's a coding efficiency, coding uniformity. And there are examples of subtle differences that you can use these sciences for all those things as well.

But this is just using thermodynamics model.

 

 [Colin Miller]

Wow. Okay. That's an impressive storyline.

What an amazing case study. For companies looking at why come to PCTS, it's, hey, we can save you a whole host of headaches or lack of a product.

 

[Michael Choi]

Yes. And unfortunately, for a lot of coding process where that precision is not required, for example, for aesthetic coding, you don't need that, right? You can live with the variation from batch to batch.

But that's not ideal. You still want to understand it and do the right thing. So I think a lot of companies kind of dismiss aesthetic coding as requiring any science, which is unfortunate.

 

 [Colin Miller] 

Yeah, no, it is. I agree with you. Changing subjects, but related actually, can you give us a low light and a highlight from your career?

I think you've almost given us one of the highlights, but let's have a low light and then a highlight.

 

[Michael Choi]

I think low light, if it covers my entire career, I would say that that low light came from hearing that this organization I led would be shut down as part of broader site consolidation. Now, this happened about three years after I left, but we put so much effort. It was very difficult because these guys worked tirelessly to elevate the site to world-class operation.

And when I went in there, they were in compliance remediation situation. And we turned it around and everything, all the metrics were going great. But seeing that their hard work overshadowed by strategic decision outside their control, that was probably the low light.

As for the highlight, this one actually has to do with the multi-billion dollar product at the time, which is the best-selling product in the world at the time. But I had this opportunity to lead a cross-functional team to troubleshoot a recurring manufacturing issue. It's one of those products where the problem was cyclic.

We would send a team in when there's a problem, and then problem would disappear. And they did a whole bunch of things. They tried to summarize what they did.

And then they go out, they get out of the project, they all get promoted. And then next year, the same thing happens again. And so this was cyclic.

So one time, they trusted me to lead a team to troubleshoot this because I was a coding expert for the company. And so they gave me a team. And it was so rewarding because I had to use every bit of engineering and modeling skill set that I had.

First, the process had to be scaled down so it can do a lot of runs. And then we had to design targeted experiments. And then we uncovered not just the process root cause, but we also determined material and operational factors that drove those cyclic failures.

So solving that problem alongside such a committed team was incredibly rewarding. So that, to me, was the highlight. And this is obviously a different product than what I mentioned earlier.

This had so much visibility because it was such a big product.

 

 [Colin Miller]

And so really, it moves to our final question of how would you speak to yourself at age 25? What advice would you give yourself now that, you know, where you are today and you were talking to yourself at the age of, say, 21 or 25?

 

[Michael Choi]

My first inclination is not to say anything because I know that I have to go through some difficult and interesting times. But I don't regret those because it also helped me build character. I think without those, I wouldn't have the character that I do.

And I say that because my generation, I had to go through some difficulties being a first generation immigrant to the U.S. As I see my kids and their kids, they're getting much better life. And I sometimes wonder if they do get well built the same characters as I have. I don't expect them to.

But then if I was to tell myself, hey, how could I have accelerated something? I would say it's absolutely okay to be different. And I should feel good about the differences that I have and celebrate them.

I think the only thing that I would caution is, hey, you also need to respect other people with different thoughts, opinions. But as long as we know the differences, there's strength in appreciating the differences together. And for me, a lot of people say I'm stubborn, but I'm not also too worried about that because I think stubbornness is useful sometimes.

It's better than being wishy-washy. But I also realize that you can't be stubborn at the expense of the people you love. What I would advise myself there is, hey, it's okay to be stubborn, but know that you're stubborn.

Admit to it when you're wrong and have a fun time laughing about it together with other folks. That's what I would say. But yeah, I think one of the greatest things about living a life that I have, I think it's a big part of it is character building.

 

[Colin Miller]

Fantastic. Wow. What a great ending there.

And Michael, this has been a fascinating discussion. And wending our way through particle coating to stubbornness is probably not two topics we need to put together. But this has been absolutely fascinating and really appreciate it.

So thank you so much. It's been a pleasure having you here today.

 

[Michael Choi]

Again, thank you so much for having me here. I can't believe how much research you guys did into actually getting these questions because you have to dig in to ask these questions. So I really appreciate that.

And that kind of speaks for you and the team that goes behind this to bring this podcast. So thank you for that. And keep it up.

 

[Colin Miller]

Thank you. We will. And thank you so much to the others that are on this podcast that helped support me.

And just for the audience, that's Amy Jarvis, Keegan Shepard, and Ashley Mauer. Thank you. Because I can't do it without them.

And again, Michael, thank you so much for being here today. Fractals is brought to you by Bracken, the professional services firm for life science and digital health organizations. Subscribe to Fractals via your preferred podcast platform.

Visit us at thebrackengroup.com or reach out directly on LinkedIn. We'll be delighted to speak with you. I'm Colin Miller, wishing you sound business and good health.

Thanks for listening.