Listen on
Overview
The promise of a new drug often stalls at the fundamental challenges of solubility, stability, and manufacturability. For biotech and pharmaceutical firms, this isn’t merely a scientific hurdle; it’s a core business impediment, directly impacting time-to-market, regulatory approval, and competitive advantage. Addressing these ‘difficult drug products’ requires a data-driven approach to formulation that balances performance, stability, and manufacturing feasibility.
Host Ross Katz speaks with Wesley Tatum, Principal Engineer at Serán BioScience, a contract development and manufacturing organization (CDMO) that tackles these complex issues using data. Tatum leads a team focused on evaluating and solving limitations in drug bioavailability and manufacturability for complex compounds. His perspective is grounded in practical, data-intensive problem-solving.
Tatum explains the role of amorphous solid dispersions in enhancing bioavailability and describes the specific data points—from crystalline and amorphous solubility to melt quench calorimetry—that guide their formulation decisions. He also highlights the importance of data standardization for building institutional knowledge and ensuring scalable processes.
Key Takeaways
Drug development success depends on foundational data about molecular properties.
Developing a reliable drug formulation requires precise data collection on crystalline solubility in biorelevant media, amorphous solubility, and physical stability metrics like the Tm/Tg ratio. This data characterizes a molecule’s fundamental limitations—e.g., thermodynamic drive for crystallization—and guides selection of appropriate formulation strategies, preventing costly downstream failures.
Amorphous Solid Dispersions (ASDs) are a key data-driven strategy for boosting bioavailability.
ASDs improve solubility by reducing the energy required to free drug molecules from a solid state, increasing the thermodynamic drive for solution. Furthermore, drug-polymer colloids formed by ASDs act as a shuttle, diffusing through the unstirred water layer and providing rapid resupply of free drug at the intestinal wall, critically enhancing absorption and predictability.
Data standardization is essential for institutional knowledge and repeatable R&D in drug development.
Standardizing data collection, processing, curation, and modeling practices is fundamental for building institutional knowledge within R&D organizations. This consistency ensures that insights from early formulation experiments are reliable and transferable, accelerating subsequent development phases and reducing the risk of errors across complex drug development lifecycles.
Balancing formulation tradeoffs requires complete data, not just maximizing one metric.
Formulating ‘difficult drug products’ demands a data-informed balance among performance (bioavailability), physical stability, and manufacturability. For instance, highly hygroscopic polymers might improve solubility but reduce tablet loading or introduce stability issues. Decisions must weigh all factors against molecule-specific data, rather than optimizing for a single metric.
Related: CorrDyn helps clients in biotech and life sciences build strong data foundations, offering services in data assessment and data engineering. We also focus on improving data quality for critical R&D decisions. To understand how better data enables drug development, see our insights on realizing data value for biotech manufacturers.
Full Transcript
Wesley Tatum: Standardization of the data collection, the data processing and curation, and the models that we’re using, those are all very standardizable and really help build that institutional knowledge.
Jason: Welcome to Data in Biotech, a podcast from CorrDyn where we explore how companies leverage data to drive innovation in life sciences. Every two weeks, we sit down with an expert from the world of biotechnology to understand how they’re using data science to solve technical challenges, streamline operations, and further innovation in their business. Here we go. Wesley Tatum, welcome to the Data in Biotech podcast.
Wesley Tatum: Hey, thanks for having me. It’s great to be here.
Ross Katz: Awesome. Just to kick us off, would you mind giving us an introduction to your background and what brought you here?
Wesley Tatum: Yeah, so I’m a principal engineer at Serán BioScience. We’re a CDMO out in Bend, Oregon, and over there I’m leading a small team of scientists where we’re evaluating different compounds, figuring out either the limitations in bioavailability or in manufacturability. My team helps figure out what the problems are, what the solutions could be, we screen the solutions, help select the lead, and scale up the processes.
Ross Katz: I think it would help if you just outlined why biotech organizations come to a CDMO like Serán and where you meet them when they come to you and then how your work kind of goes from there.
Wesley Tatum: Yeah, so there’s a lot of different points in the process of drug development that they can come to us. But by and large, they come to us because we have an expertise in formulation. Formulation and manufacturing are kind of the last two bits before you’re putting drugs into people. So they’re coming to us, let’s say on the early end, it could be discovery, they have a couple of different candidate molecules and they want to compare them, try to understand what’s going to have the best exposure, what’s going to have the most stability, the most manufacturability. We help them really early in preclinical studies, narrow down those compounds. Maybe they already have a compound lead that’s selected and they need help with GLP Tox manufacturing, we can help them with that. Or they’ve already gone through GLP Tox and they need to develop a first-in-human formulation. A lot of your Tox formulations are suspensions and solutions, and typically you’re wanting an oral solid dosage form, whether that’s a tablet or a capsule. We can take their drug and do that and compare the exposure back to their GLP Tox exposures. Or on the further end of things, maybe they’ve already gone through phase one, phase two. They were maybe pushing through a capsule, like a dry powder in capsule type formulation, and they needed something that’s a lot more scalable. That’s the other thing that we’ll do is we’ll take in phase one, phase two programs that need to be reformulated for either bioavailability or manufacturability. So all along the development roadmap they can come to us because of our expertise.
Ross Katz: Yeah, when we talked prior to the podcast, you all mentioned that Serán specializes in difficult drug products. Would you mind just telling us what are the kinds of challenges that make a drug product difficult? And then how does that alter the approach that Serán takes when working with organizations that have difficult drug products?
Wesley Tatum: Yeah. The fun part of it is that difficult drug products are difficult for any number of reasons. Especially with the new molecular space, the compounds are lower solubility, they’re more lipophilic, but not lipophilic enough to enable say a lipid liquid-filled capsule or something. Maybe it’s just solubility or maybe it’s permeability. Really they’re coming to us because we like to take a technology-agnostic approach. We do a lot of spray-dried dispersions, but that’s by no means the only thing we’re going to try. We’re going to try HME is another option or just a conventional crystalline tablet formulation with maybe micronization or maybe you need an acidulant or precipitation enhancer. So really what we’re doing and I think our expertise is trying to understand really what’s limiting the molecule and how we can come to a robust process that’s scale up.
Ross Katz: Yeah, that makes a lot of sense. So what would interest me is an organization comes to you with a new molecule and a new compound, and they want to understand from you how to improve the solubility or the permeability or the manufacturability of the compound that they’re bringing in. I’m interested in what types of data do you collect or how do you characterize where you take the conversation from there.
Wesley Tatum: Yeah, especially if they don’t have any data that we can leverage. I guess another really common problem statement is they had all this data when they were first developing the drug substance, and now they have a GMP drug substance manufacturing process, all of a sudden they’ve reduced their impurities, they’re getting different crystal forms or higher levels of crystallinity, and all that data might be totally unrepresentative of the material that we have to work with. The big data that we want to take to understand what the problems are going to be are solubilities. Crystalline solubility in different bio-relevant media. That’s fasted state gastric fluid, fasted state intestinal fluid, the base buffers that go into those. We’re really trying to understand how the drug is thermodynamically soluble throughout the GI tract. And then we also want to understand the amorphous solubility. If solubility is going to be the limiting factor here, we want to understand how much of a gain just going to the amorphous form can get you. We have a couple of experimental protocols to be able to take those amorphous solubilities. And then if we or they or us together have identified that it needs to be amorphous, we also have some screening methods that we can use for looking at different polymers that it might be most compatible with. Solubility is really important, but we also want to understand a little bit about the nature of the molecule and the crystalline drug substance itself. Melt quench is a really common technique on differential scanning calorimetry. Essentially you take your crystalline material, you heat it up until it melts and you quench it down and capture it in the amorphous form, and then you heat it up again and you can look at the melting point versus the glass transition temperature. Where those are individually can tell you a little bit about physical stability, different types of processes you want to turn to. But then the TM over TG, that ratio also gives you insight into the thermodynamic drive for crystallization. So if we are going to do an amorphous form, how much are we going to be able to push the drug loading, which might be purely a factor of how stable we can get it versus performance. That’s the main thing is what are the physical stability risks and what’s going to be the tendency to recrystallize.
Ross Katz: So you mentioned solubility and you mentioned amorphous forms. Would you mind introducing people to amorphous solid dispersions and why amorphous solid dispersions help with solubility and what you’re sort of driving at with the experiments that you’re doing in order to characterize whether that’s the right thing or whether there are other formulations that are going to be more appropriate.
Wesley Tatum: Yeah, absolutely. Just starting with your first question, why an amorphous form and what is it? When you have your crystalline structure, there’s kind of two steps that it takes to solubilize a drug. First is your enthalpy of sublimation, essentially. You have to break that molecule out of the crystal lattice and there’s an energy to do so. And then there’s an entropy of mixing. You have that drug and now you need to get it off and be free drug before it, because it wants to be in that more stable state of being in the crystal lattice. What an amorphous system does is it reduces that first step. Now all of a sudden you don’t have these really stable crystalline structures and you have hydrogen bonds, maybe dipole-dipole interactions, all these positive interactions that make crystal lattices so stable and you disrupt those. Now maybe you only have some local interaction between them, and it’s a lot lower energy cost to be able to remove that drug from the whole solid and then your entropy of mixing is going to be the same. You’re reducing energetically how much it takes to solubilize the drug and therefore there’s a higher thermodynamic drive to have drug in solution. Your crystalline solubility is always going to be lower than your amorphous solubility. And when I’m talking about solubilities here, I’m referring to only free drug, not drug in micelles or drug in polymers or colloids, which is another aspect of amorphous dispersions. That’s just if we’re looking at the crystalline drug versus the amorphous drug, that’s the basis there. Now an amorphous solid dispersion is then when you take a polymer and you mix it in, and that’s going to further separate the molecules and it’s going to help make them a little more stable. Because if you just have the amorphous drug itself, it’s going to want to tend to go back to that crystalline state, whether it’s you’re sitting in your GMP stability experiment at 40/75 and you got heat, you got humidity, and it’s always going to want to go back to that crystalline state. So you can stabilize it by dissolving in a polymer there. And then the polymers also have implications around bioavailability as well, but that’s a different aspect of it.
Ross Katz: I’m really interested if you can connect the formulation that you’re doing in terms of how your clients are thinking about dosing or thinking about the way that the drug is metabolized by patients. I think that entire logic cycle is really interesting if you’re able to.
Wesley Tatum: Yeah, happy to do so. In addition to having higher free drug just because you have the amorphous form, you are also able to form drug-polymer colloids. There’s a lot of really good literature out there, some really good research being done on how those form, why they’re bio-available. The high-level summary is you have drug in colloids and those colloids themselves are able to also diffuse. We’re actually presenting some research this week that characterizes some of the different colloidal properties. Z average, hydrodynamic radius. We can also get at density through centrifugation. And we’re trying to understand how different polymers, different drug loadings, how those different colloids differ. Regardless of how they differ, what they’re going to do is they’re going to diffuse, they’re able to also diffuse through the unstirred water layer. That can be really limiting for some drugs, especially if they’re not as lipophilic, they’re just hydrophobic. They’re going to get to the unstirred water layer and they’re going to take a really long time to diffuse through that to get to your intestinal wall. But you can have drug-polymer colloids acting as a shuttle. They’ll go across the unstirred water layer, they’re going to just hang out there at the intestinal wall, and as your drug gets absorbed, you have free drug right there ready to resupply it. Rapid resupply is how we refer to it and it can be incredibly enabling for bioavailability.
Ross Katz: Yeah, interesting. The goal if I’m understanding correctly is to make it so that the drug is being taken up by the patient’s system at the right rate and in a predictable way, make it so that it’s something that can be manufactured easily and at scale, and these are the different components that you’re trying to balance as you’re doing these kinds of experiments.
Wesley Tatum: Exactly right. Yeah, I think balance is the key word there. When we’re just doing our screening studies, we don’t want to just maximize performance at the cost of stability or at the cost of downstream manufacturability. The polymers that you choose will have different hygroscopicity is a big one. PVP-VA, HPMCE3 are really common amorphous solid dispersion polymers, but they’re really hygroscopic. So you can fit less of the ASD into the tablet. They’re going to have different stability issues compared to maybe an enteric polymer like HPMCAS or Eudragit. Those are going to be able to enable higher ASD loading into the tablet and they’ll have less water uptake, but then you also have to balance stability risks because they have a lot of acid content in them. So if you have a drug that degrades in the presence of acid, it all goes back to balance. What’s going to be best for all of these different pillars of formulation, not just bioavailability.
Ross Katz: You mentioned rate-limiting steps. You’re trying to understand what’s blocking the drug’s moving forward. Is it performance or is it manufacturability and then what are the issues with performance or manufacturability that you have to resolve in the process of developing the formulation or the way that it’s being manufactured. Am I thinking about that right?
Wesley Tatum: No, that’s exactly right, yeah. You’ve got it.
Ross Katz: You’ve got all of these ideas, some of which you’ve just shared of the decision tree of potential formulation decisions that you could make, different polymers, different excipients that you could bring in to resolve different issues with the drug and make sure that it’s manufacturable, that it’s soluble, that it’s stable. How do you narrow the process? How do you determine what experiments you need to run or how you get to the point where you can then make a proposal to a customer about the appropriate formulation?
Wesley Tatum: Yeah, we actually do something called a developability assessment where we’re taking that solubility data and thermal data and we plug it into a couple of different models. BCS, DCS, those are fairly well known, biopharmaceutical classification system, developability classification system. We’re just trying to bin these different problem statements. How much are we focused on just solubility, how much can we get away with improving just dissolution rate. The difference between DCS 2A and 2B is low solubility, but in 2A if you just help it dissolve faster you’ll be fine. You’ve got high enough permeability. That’s going to drive which technologies we might evaluate. We’re also looking at the thermodynamic rate-limiting steps for bioavailability. Kiyohiko Sugano’s put together a really good framework called the fraction absorbed classification system where you are looking at those three rate-limiting steps. Dissolution, solubility, and permeability. Just by looking at the molecular properties, some of these thermodynamic measurements that we’re making, we’re trying to understand how much are we going to be solubility limited versus permeability limited. And then one step further, how much of that is at the unstirred water layer boundary versus at the intestinal epithelium boundary. The first thing we do is we take those measurements, we put it through these different models, and we understand what technologies are even worth exploring. We’re not going to just say, oh, micronization will help if you’re BCS class four or 2B. Clearly you need some additional solubility enhancement. There is that broad decision tree and mostly what we’re trying to do is prune off forks of that, and each of those decision trees have a fractal amount of other variations in them.
Ross Katz: What are the conversation points with your customers as you’re going through this process? How do you communicate how that decision tree is evolving or which aspects you’re thinking about pruning and how do you gather input from your customers about what you should try or not try.
Wesley Tatum: Yeah. It is constant conversation and that’s something that we also really like to pride ourselves on is we’re not just taking their molecule and running with it. It’s their molecule and we’re just a part of the team. We’re meeting with them weekly, bi-weekly. Maybe we’re in the middle of something and we get some really interesting data or some really concerning data, we want to take that to them immediately, talk through why we think it’s promising, why it’s limiting, and possible ways around that. The developability assessment, the report, that provides a really good framework for where we think the thermodynamic limitations are, and we can give caveats to each of those. Like, oh, this says you won’t get this, but really there’s all these other factors to consider, so this might still be a beneficial avenue to pursue. Or no, this seems right out. I don’t think we can pursue that. Maybe they still want to pursue it, so we’ll give it our best effort. But we’re trying to take into account what kind of manufacturing processes that they would like to pursue. Maybe they really don’t want to do spray drying for one reason or another, there’s more unit operations, you’re using solvents, and they really want to focus on HME. Okay, well how’s their thermal stability? How strong are the interactions with the polymers? Is that a feasible route? Do they have the material for us to look at an HME? You need at least a half a kilogram of API to get on something at scale. There’s smaller things you can do, but there are material considerations. We’re balancing maybe they only have a gram of material to work with. Okay, well let’s evaluate these avenues first and then we can come back and re-evaluate these other avenues later. Or maybe they have a really high dose. With some of these new compounds, it’s not unheard of to have a gram or more of active that you need to dose in a day. And that’s really going to drive what sort of technologies you think are feasible.
Ross Katz: Yeah, it’s interesting. You’re sort of helping them navigate the scientific constraints of the problem while also sussing out the economic constraints and the supply chain constraints and the logistical constraints, the engineering constraints that are on the problem at large. While you can communicate the scientific and physical and chemical nature of the way that the drug needs to be managed, you also have to track all of the external things that lead to the drug being appropriately manufactured.
Wesley Tatum: Exactly right. Yeah, there’s a lot of just practical constraints. Let’s say we’re doing a formulation screen and we have one polymer that just knocks it out of the park. We actually have this quite often where a 25% drug loading in PVP-VA looks really good in vivo and in vitro, and let’s say maybe HPMCAS is good, it’s better than the amorphous form, better than the crystalline form, but it’s maybe 30% less good than the PVP-VA formulation. And the client might say, okay, we have to do this PVP-VA formulation. That’s going to be our one. Well, we can take them through the math and show them that we can fit 30% more of the HPMCAS ASD into a tablet, and so really the bioavailability all levels out. It’s taking those practical constraints and all those downstream considerations and trying to think of them as early as possible so that we can make the right decisions at the start. We don’t end up painting ourselves into a corner.
Ross Katz: Yeah, that makes a lot of sense, and it’s also the point where picking the right partner early on that can take you through the entire journey helps you to make good decisions early on that lead you to better situations later on.
Wesley Tatum: Exactly. Yeah, we have a non-zero amount of what we call rescue programs where there was a formulation that was landed on for maybe phase one and the client recognizes it’s not scalable and they come to us. We’re trying to figure out, based on how they got there and all the data that supported that, what can we figure out, what additional data do we need to be generating to come to a more robust process.
Ross Katz: Yeah. Each of these screens that you’re talking about, these assessments that you’re talking about, I’m interested in how standardized is the process for each of these and how much of this is you coming up with methodologies on the fly given the tools that are in your toolkit, and are there consistent computational approaches that you’re using that scaffold these things and allow you to make them more repeatable or make them more efficient, that sort of thing.
Wesley Tatum: Yeah. That’s a great question. I would say it gets more tailored to the compound as you go through the process. When we don’t know anything about the compound, we got to get the basic data. DSC, PXRD, aqueous solubility as crystal component, aqueous solubility as an amorphous compound. Those are pretty standardized. The gastrointestinal tract’s going to be pretty consistent, so we’re always looking at those. Now the big thing is that we’re trying to make sure our experimental protocols are also very standardized because some of these compounds can be incredibly sensitive to ionic strength. If you have one analyst that overshot the pH of the media and they had to titrate it back down, and your ionic strength is instead of 28 millimolar for FaSSIF V1, you’re looking at more like a 35. Maybe it doesn’t matter, or maybe it really does. You don’t know what you don’t know is the big thing. We’re really trying to have standardized experimental protocols so that we can tailor what we’re actually putting into those experiments to the compound itself.
Ross Katz: Yeah. I’m interested in moving from compound to compound, from customer to customer. How transferable is the knowledge that you get? Is the extent of the transferability just the experience of going through the process multiple times with multiple customers, or are there aspects of experimental design or of data gathering or of computational approaches or of standard operating procedures or just the way that you work with the compounds that you’re receiving that is standardizable or transferable for the organization’s knowledge to grow over time.
Wesley Tatum: Yeah. I guess I’ll first say that even within a molecular series, let’s say a client brings us multiple compounds. You’d be surprised how often a methyl group or changing the site of your chlorination can drastically change all of your molecular properties. To some extent, the specific knowledge of classes of compounds is not that transferable. But like you’re alluding to, the protocols, the pitfalls, all of that experience that you get and the problems that come up that you have to solve, those are incredibly transferable. Standardization of the data collection, the data processing and curation, and the models that we’re using, those are all very standardizable and really help build that institutional knowledge.
Ross Katz: Are there best practices that you’ve found for ensuring that your team is learning as it’s growing, is learning from the things that have come up in projects previously, or how do you think about embedding that knowledge in the organization as a whole?
Wesley Tatum: Yeah, that’s some of the stuff that’s hard to put into an SOP or a work instruction. What we’re doing, and a big part of my role is having seen so many of these programs and having been in the lab myself, is being able to look at the data and understand where those different pitfalls might be or how we need to adjust an experiment to be able to tailor it to the compound. With my team, a lot of it is training them by experience. Okay, we have this data, we have these trends that we’re seeing, what do you think of that? Just bringing them through and having them understand.
Ross Katz: As with all science, you’re always up against the epistemic uncertainty of you’re never going to have 100% confidence in whatever conclusion you’re drawing. But then you’re also dealing with organizations that are under time constraints, material constraints, budget constraints in addition to the epistemic constraints of how can we even know what we’re trying to discover here? I’m just interested in how do you determine what questions need to be asked and are there situations where you need to push back on the customer to say, actually we’re not going to reach a level of comfort with the conclusions that we’re going to be able to draw at this level of budget or that sort of thing.
Wesley Tatum: Right. Yeah, I think we have those conversations a lot, both internally and externally. What we do is we have all of the questions. You can list them out, you can make a list. And then the most important thing is to rank order them and say, okay, what is critical path, what is non-critical path. What are the need-to-knows to advance and to hit your IND submission date or your GMP manufacture. What do we need to know to get there versus what is a nice-to-know that we can tack in along the way. If we’re going to be doing a screening study, let’s say we already have a formulation and we’re about to do a demonstration batch, and we want to understand something about the stability or different excipients. Okay, well what of that can we tack in to this work that’s already there? How do we make it so that it’s not its own standalone experiment that’s gating to the rest of the program? How much can we find synergy in what is critical with what we want to know.
Ross Katz: I love that answer. Do you have an example of a company that you’ve worked with that came to you with a particular compound and how you determined which questions you wanted to keep in and which ones you decided to leave out?
Wesley Tatum: Yeah, I think probably this most commonly comes up with bio-relevant dissolution. It’s a fantastic tool, it can give you great mechanistic insights into things. But it’s not predictive. It’s biomimetic-ish, but you’re glossing over a lot of differences between vials or USP apparatuses and the human gut. One common thing that we get caught in is okay, why do we see these differences, either experiment to experiment or between formulations? And it’s really tempting to just deep dive into okay, maybe there’s something different in the process for the manufacturing, you can even get down to lot-to-lot inconsistencies of excipients. It’s really easy to fall into that rabbit hole. Sometimes it’s just easier to put it in vivo. Dissolution is cheaper and it can be quicker and we want to leverage it as much as possible. But at the end of the day, you can get a concrete answer by putting it into an in vivo study. So that’s probably the most common thing that happens. And then I’m trying to think about an example of synergy that we can find. One of the problems that I alluded to is drug substances are evolving as we’re trying to develop drug products. We’re doing a manufacture with maybe a medchem lot, it’s only a couple grams, maybe only a couple hundred milligrams. We can time things or if the timing works out and they get new drug substance, we can run that in as a control. Whenever we’re doing bio-relevant dissolution, there can be variability from experiment to experiment, so we like to always include a crystalline API control so we can see, just a sanity check: Did we make the media right, did we forget to turn on the heater, are we stirring at the same speed? All of those can have effects on the curves that you’re seeing. We have our crystalline API, and then we can always take additional lots, or maybe we’ve done a new round of formulation screening where we’re pushing a drug loading. We have our polymer, we’re pushing drug loading. We can take, at the same time they’re doing maybe suspension development work for GLP Tox. What we can do is we can tie in some of that suspension stability work with our new round of formulation screening. If we take our suspension and we age it for 24 hours or a week, then we can just put that into the dissolution study that we’re already running and derisk that as early as possible.
Ross Katz: Yes. It just strikes me that you’re working with these companies that are very much in motion, that are in their own external discovery processes even as they are asking you to take on this discovery process on their behalf. You also have to be responsive to the changes that they’re making on their end and how their drug is evolving or new constraints that are being applied or new knowledge they have about how it needs to be dosed or how quickly it needs to dissolve or things like that. Yeah.
Wesley Tatum: Right. Yeah, and the dose always creeps up. Yeah, it’s a constant conversation and we love to tell the clients, the more information you can give us, the better data we can give you, the better formulation decisions we can make together. As much as we try to be open and transparent and responsive, the best relationships are with clients that are equally as open and transparent and responsive.
Ross Katz: Yeah, interesting. What does exploratory R&D look like from the Serán CDMO perspective? Are you all doing internal R&D in addition to the R&D that you’re doing on behalf of your clients? And if so, what does that look like and how does that support the work that you’re doing for your clients?
Wesley Tatum: Yeah, so we have three main branches of R&D. One is developing new technologies, whether that’s a new manufacturing process or an augmentation of an existing process. We also have another branch that is continuous improvements. We have different design spaces, how can we improve them, how can we make them more robust, what sorts of new studies can we tack on or include to derisk things or how can we better design our standard experiments to give us what we’re looking for. And then there’s a more fundamental side, and this is where me and my group tend to fall in more, is going through all of these different compounds and sometimes you get textbook data, you can make clear decisions. Sometimes it’s a lot more ambiguous and you start to see things that you wouldn’t have expected, you wouldn’t have predicted. Okay, how can we drill down into this and figure it out? When it comes to those kinds of questions, there’s two approaches: One is talk to the client and say, hey, we’d love to do these experiments, we think they’d be interesting scientifically and they could help answer some of these questions if they’re not critical path. Then there’s discussions about when and where and how. The other option is find model compounds. This is difficult because a lot of the model compounds are over a decade old or more. They’re from a different paradigm of molecules than we’re seeing come through the pipeline now. We’re seeing such different approaches to the chemistry and the targets themselves are evolving, so it can be really hard to find a model compound that replicates the behavior that you’re seeing in this client’s compound. Some of it’s a conversation around if we were to do this for you for free, could we include it in a poster presentation with all the data blinded? Or if they’re willing, we could publish. But there’s a lot of IP considerations around that. So that doesn’t happen that much.
Ross Katz: Yeah, that makes a lot of sense and also I didn’t think about the fact that the model compounds are also lagging behind the biotech ecosystem because if you’re able to synthesize a compound you’re probably focused on the patients and the use case and the actual launching and not the modeling that’s going to allow for studying manufacturability or formulation or things along that line. Am I understanding about that right?
Wesley Tatum: Yeah, exactly. And just how they’ll interact with the polymers is different as well. The typical paradigm with amorphous solid dispersions is spring and parachute. You have your amorphous form and you have faster dissolution, higher solubility, so it springs up and then it crystallizes out and your polymer’s trying to just extend that time before it precipitates out. But compounds now can completely self-sustain their own amorphous solubilities. We do still see some of the rapid crystallizers, but we see a lot of other behaviors as well. Those are often the ones that are more interesting, more salient from a research perspective. But those are the ones that are going to be harder to replicate.
Ross Katz: It strikes me that in the absence of being able to predictive advance the types of compounds that are going to be brought to you, it’s hard to do exploratory R&D on those compounds. The model compounds have to be in the same modality space as the modalities that are being brought to you. Am I thinking about that right?
Wesley Tatum: Yeah, exactly. How we’ve been approaching it, we can either try to have something that’s as similar to the compounds that we’re seeing as possible, or we can just have design of experiments: We’re only going to vary one thing at a time, okay, these two will control for molecular weight, these two for PKA, weak acid versus weak base, log D. There’s all these different physico-chemical properties that if you look at enough compounds you can start to build out an experiment to compare them all. But then you have to find the time to do more.
Ross Katz: There’s a lot of talk in the external world about AI accelerating molecule discovery, trial design, I’m assuming drug product and formulation development. But it would be interesting hearing from you, where are you using AI, where do you see the opportunity for AI applications in what you’re doing, and where is the external perspective on AI automating all of this a bit off base.
Wesley Tatum: No, it’s a great question, and I know it’s something that we’re all really interested in ourselves internally. How can we take all this data that we’re generating and have it feed back into future programs? I guess I’ll say the biggest limiting factor is data. If we’re going to use something like a neural network, or some sort of Bayesian optimized model, you need tons of data. We’re not talking about 50 data points or 500, it’s like a million data points before you have a model that’s really fine-tuned. Even though our company’s 10 years old, we’ve worked with hundreds of compounds, we just don’t have that level of data. The other thing with data is since we’re taking all of our other clients’ data, there has to be legal agreements around what we can use their data for, what we can’t. That can also be a limiting factor. And then finally, the final thing is this data curation that we’re talking about. The drug substance is evolving, our experimental methods are evolving, how we’re processing it is evolving. Even though we might have data that’s 10 years old, we might not be able to include it into our data sets for our compounds today. We’re really interested in it and we’re working on standardizing a lot of these things not just for translatability, teachability, having new people come in and be able to learn and produce data exactly like we’re doing, but also so that we can build out these data sets so we can at least do trends. If we want to do AI and ML, we need to make sure that we can trust our data. That’s also going to guide what sorts of models we’d want to use. With this level of data that we’re producing, maybe we’d be better suited for like a XGBoost kind of model. The salient part of those as well is you can do much more sensitivity analysis and see, oh, wow, I didn’t realize that this physico-chemical property was going to be so governing of which formulation we ended up finding was the best in vivo.
Ross Katz: Yeah, that’s really interesting. It also brings up the question of automation. Obviously one of the sources of the variation that you’re describing is human hands touching the experiments that you’re doing. Interested in where you’re seeing automation being applied most effectively internally and then where the limits of where lab automation can be utilized to do the kind of work that you do.
Wesley Tatum: Yeah, media prep is probably a big thing that needs to be standardized and probably could be automated. I think HPLC data workup and analysis, that’s another big one that we can at least make semi-automated if not fully. Agilent’s got a nice robotic arm at its booth that can help out with that and in-situ dissolution as well. If we’re thinking about it from the QC side and release tests, those are much more standardized and automatable processes. There is definitely untread ground for things that can be automated. Where it stands today, I think the best we can hope for is standardization. But I would love to see bio-relevant dissolution become something that’s a lot more well-understood and automatable.
Ross Katz: Yeah, that makes a lot of sense, but also standardization is the precursor to automation. In the absence of standardization, you can’t automate the thing, otherwise you’re doing something different. Do you see in the next, I don’t know, two to five years, that generative or predictive models are going to fundamentally change the kinds of formulation science that you’re doing?
Wesley Tatum: Yeah, I think two to five years it’ll probably be fairly similar to what it is now as we’re building up standardized data sets. But I think that’s also the transition time to where we’ll start to be able to include a lot of these smaller data models. We’re already incorporating a lot more first principles models into our workflows. And that’s a slippery slope to AI and ML. I think there’s also some really cool work being done around physically informed AI models. Your loss function can include some of these first principles calculations that we’re doing. Even though it’s not necessarily a first principles calculation, it can still be informed and you can guide these generative AI models with your fundamental equations.
Ross Katz: Yeah. That’s what I see a lot, especially from generative protein design type companies, is they’re doing a lot of that feeding back first principles, physics-based models into generative models and then trying to constrain it as much as possible and then only testing when absolutely necessary, and that’s the trends out there.
Wesley Tatum: Exactly. You touched on something there: it is a feedback loop between the physics-based and the generative and the data. As you kickstart the cycle, things will refine and then they’re only going to keep on feeding into each other more and more.
Ross Katz: I’m glad to hear you verify that because that’s my perception of how it’s working. You’re at the intersection of so many different stakeholder groups with so many different, highly specific expertise, all of which needs to feed into the decisions that are being made collectively between you and the client about which experiments to run, how to invest the resources, how to optimize the formulation. Can you share anything that you’ve learned about how to communicate across people whose expertise is chemistry or biology to people whose expertise is engineering to people who are more on the commercial side to people who are on the manufacturing side to people who are in the lab hands-on experiments and may not have complete knowledge, or any stakeholders that have partial knowledge of some of these domains but not full knowledge? It just seems like the coordination problem has to be a big one there. I’m just curious what lessons you’ve learned about how to tackle that.
Wesley Tatum: Yeah, that’s actually one of the aspects of this job that I didn’t realize was going to be an aspect of the job, but also that I’ve grown to really, really enjoy. I think a big part of it is learning each other’s vocabulary. A chemist and an engineer might be using solubility differently versus formulations where we’re talking about crystalline versus amorphous versus colloidal versus all these different solubilities. That really is at the heart of it: you have to understand what they mean by the words they’re saying, even if we’re using the same words. But I think everybody speaks risk. When you’re talking about the data, where it could take you, and the concerns that I have, putting together a table that gives, here’s a decision, here’s a rationale, here’s a formulation choice, here’s why we’re choosing that. If you can just break it down into the categories of why you’re saying it and why it’s important, I think that’s a big thing. And then also understanding, and this is where it really comes back to building relationship with the client, understanding what sorts of problems they’ve had in the past, what scar tissue do they have around different technologies or avenues that we might want to explore, because that also is going to inform how I communicate these different things to them. I need to understand what they like, what they don’t like, and how they’re going to talk about it.
Ross Katz: Yeah, it just strikes me that whenever you’re providing a very technical service on someone’s behalf, you have to understand—you mentioned the term vocabulary—you have to understand the terms under which they understand their problem before you can help them to understand their problem any more effectively. Because if you’re coming at it from a completely different and disjointed methodology and way of speaking, even if you’re giving them really good advice or really good guidance on how they should be doing their formulation for example, they’re never going to be able to take in that advice and actually understand the rationale behind it.
Wesley Tatum: Exactly. And a picture’s worth a thousand words. We find ourselves very often turning to PowerPoint. We’re always presenting data in PowerPoint, so you can make diagrams in there or use something else if you need to get more fancy. But being able to just distill it down to something visual so that then you can say, okay, here’s when I say super saturation, here’s what I’m saying versus how I think you’re using it.
Ross Katz: Yeah, that’s interesting and also PowerPoint gets such a bad rap. Scientists are always putting things in PowerPoint, but what you’re sharing is that actually the reason to take it out of the spreadsheet or out of the database or out of the experiment, out of the LIMS system and into the PowerPoint is for the purpose of translating it to your stakeholders and meeting them where they are with, which may not be standardized in terms of how an AI might read it, but it is important for humanizing the data and making it consumable.
Wesley Tatum: Exactly. Another layer to that is we’re oftentimes interfacing with part of a team, and they have to turn around and explain our data to the rest of their team. We wouldn’t be able to do that without having well-designed slides, really in-depth conversations around the data and our conclusions, and then making sure that those are captured in the slide so that they can turn around and present it. Actually a lot of our legacy data is you’re having to go through different Excel spreadsheets and PowerPoints, figure out where it was consolidated, go back to the lab notebooks, figure out how it was collected, and we’re going through right now and trying to consolidate it into machine-readable forms so that we can then, and that helps you standardize the way you take future data as well as you understand what is not useful for machine-readable.
Ross Katz: Yeah. As I’m hearing you discuss it, it also strikes me how iterative this process is, that you learn about how your data model needs to look by the questions that people ask you and by the slides that you need to develop, but you can’t know that necessarily in advance when you’re running the experiment. It’s just something you learn over time and there’s a feedback loop between your interface with your clients and the way you design your internal systems to capture and log that data. Right? And yeah, exactly. In terms of internal systems, what do you think is the next stage of maturity for you all in developing especially data infrastructure, but any internal systems that you feel need to be in place for you all to be able to do your work more efficiently, more effectively, or be able to tackle new classes of problems that clients are bringing you?
Wesley Tatum: Yeah, I think electronic lab notebooks. We’ve made that transition almost fully in development. We’re going to be transitioning throughout the company. It not only standardizes how you write your protocols, it just makes the data that much easier to go back to and collect. A lot of the databases that the ELNs are stored on, you can expose them and do SQL and NoSQL queries and that is an immediate boon. We’ve made that transition this year. I’m really excited to start digging into the data.
Ross Katz: Can you share a little bit about what that transition was like? What were the hardest parts of transitioning into ELNs—and maybe before that, how did you all decide that it was time, and then after you took it on, what were the hardest parts of making it happen?
Wesley Tatum: Yeah, we decided to take it on I think out of, it helps with review, it helps with standardization. Those were really the catalyzing factors. But we also have people at our company that have helped set up ELNs or came from labs that only used ELNs and so there’s just already a lot of interest in it. We’ve been working on it for a while. The hardest part is building out the protocols. Especially in a development environment where we are tailoring our experiments in so many ways and we have such variable numbers of samples that can go into it and so many different experiments and unit operations that we’re running, building out all of those templates takes a really long time. It’s a lot of working with the ELN company to try to see what they already have templated, what you can adapt of those, what you need to write from scratch, and then there’s a lot of learning how to write them from scratch. I would say writing the protocols in a way that fits our system is the majority of the time that it’s taken.
Ross Katz: What is the hard part about writing the protocols? Translating it into the ELN can understand or is it documenting things that have never been documented before? Is it hashing out among each other what the protocol should actually be versus what we’ve actually been doing under something else, I don’t know.
Wesley Tatum: A lot of it is edge cases. You start out by building your protocol for the standard, okay, it’s a bio-relevant dissolution, five samples in of two, we’re going from pH 1.6 to 6.5 with bile salts. Okay, but what if we’re doing a bio-relevant dissolution and trying to derisk food effects or we’re trying to understand drug-drug interactions where the patient population is going to be taking proton pump inhibitors? Or maybe they have a physiological achlorhydria where their stomach pH is going to be higher. So the media right there, you’re having to change or develop additional recipes. Okay, well what if we want to only do a one-stage dissolution? Or what if we want to do a three-stage dissolution either because we’re doing a colonic pH or because we want to understand how our undissolved solids or precipitated solids are going to redissolve and get some insight into what’s going on into the intestine that way. And that’s just dissolution. And then there’s manufacturing processes of okay, how many samples do we need, how many different formulations do we need to produce? How many samples do we need to take? What types of samples? Where are they going to go? How do we track all of these? Is a stability sample the same thing as a dissolution sample? It really is just taking all of these standardized Word doc protocols that we adapt to every program to every step of the way and trying to get a single document or protocol that captures at least 90% of those and being okay with going back and building specialized ones.
Ross Katz: Right. Yeah, so it’s just like an extension of all the different questions that you might ask times all the different permutations of the ways that you might ask it, and you end up with just an endless list of protocols. As we head toward the end, as you think about your experience, so many organizations, like Serán, you’re a product of your history and the way that you grow and develop, the clients that you take on early on and the methods that you’ve developed and all of that. I’m interested, if you were designing a data-first CDMO from scratch, are there lessons learned or ways that you would think about it differently if you were starting again?
Wesley Tatum: I think starting with an ELN is the big thing, and then that way another important thing is you can track how your experiments evolve. Because experiments are going to evolve, science is evolving, the models are evolving, and so how you collect the data is going to change over time. But if you have it that way in an ELN, everything is going to be tracked. I think that’s the big thing is thinking about how you’re storing your data. File tree structure, that’s another big one. Oftentimes those evolve organically. Yeah, but if you can build something to crawl through your file tree and pull out your different bits of data, that would be huge too. Yeah, I think those are the two big ones.
Ross Katz: Where can people go to learn about you and the work you do at Serán?
Wesley Tatum: We have a learning center on the Serán Bio webpage. All of our white papers, any of our presentations. You can listen to them all. And then any white papers are going to be there as well.
Ross Katz: Yeah, well awesome. It’s been great to see you here at AAPS PharmSci 360. Really appreciate the time and look forward to connecting down the line.
Wesley Tatum: Yeah, you too.
Jason: And that’s it for this episode of Data in Biotech. If you enjoyed the episode, please subscribe, rate, or leave a review in your podcast platform of choice. See you next time.
