How Modeling Helps Make Better Drugs, Quicker

From target product profile through clinical proof-of-concept, there is a better way to approach drug R&D

Model-based drug development offers a faster, more efficient way to identify a molecule with strong potential to serve as a medicine, assess clinical performance, and demonstrate safety and efficacy to the U.S. Food and Drug Administration (FDA).

Okay, that sounds good, but what does it really mean? Think of model-based drug development like a blueprint for a house. Sure, people have for centuries built shelters, from yurts to cabins, without architectural plans. But if you’re trying to build a house, the seat-of-the-pants approach doesn’t work- you need a blueprint. The blueprint allows for more complex designs, and wastes a lot less money in the end. Extending that metaphor, if we want to achieve challenging goals with our candidate drugs —expand cancer survival rates, relieve symptoms of autoimmune disease, slow neurodegenerative conditions, and so on—model-based drug development can serve as our blueprint. It can help us direct research dollars and time in a more targeted, efficient, and ultimately effective manner.

Resources are always limited. Doing more with less, therefore, empowers us to achieve more for patients, sooner. And for biotech companies, it means creating better products and beating competitors to market along the way.

That’s the promise of model-based drug development. But what does that translate to in practice, for pharmaceutical research? To answer that question, let’s review the drug development process in brief and see where model-based drug development fits at each stage.

Step 1: Identify the target product profile (TPP)

The first step in drug discovery is determining what to look for in the first place. What are the characteristics of the new medicine that, hopefully, lies at the end of the long and winding road through pharmaceutical research and clinical trials?

Traditionally, a target product profile (TPP) has mostly amounted to a market assessment. Teams evaluate competitor products and compile a list of features that would make for a “better” medicine.

The problem is that when approaching this question from the standpoint of a marketer, it’s easy to engage in “pie in the sky” thinking. If a competitor drug must be taken three times per day, why not look for a once-daily or even once-weekly pill? Better yet, if the other guy’s product is an injectable, how about an oral treatment with the same, or better, clinical effects? Patients don’t much like getting shots, after all. It’s easy to pull a TPP out of thin air!

Okay, maybe that’s a bit of an exaggeration, but the point is this: a market-centric approach will frequently overlook whether the TPP is realistic. If we’re honest with ourselves, we will accept that the underlying chemistry involved in medicines, not to mention how they interact with the human body—the pharmacokinetics (PK) and pharmacodynamics (PD)—do not always cooperate with our human aspirations. In other words, the product we’d love to create may be an impossibility, within the limits of the available chemical matter or biological mechanism.

Model-based drug development approaches to pressure-test the idea involve building a model of the system (usually from the competitor’s clinical data) and then asking the question “how much would we have to improve on the competitor’s molecule to get a commercially/clinically significant outcome?” Thus, modeling is a tool that provides insights into feasibility early in drug discovery. For a novel mechanism of action, a model-based approach can also be valuable. To give just one example, before starting an antibody-drug conjugate project, it’s worth asking the question: “given target abundance and turnover, antibody PK and warhead potency, how stable does my linker have to be”?

Pressure testing your TPP before investing many thousands (or millions) of dollars trying to boil the ocean…well, that just makes good sense.

Step 2: Design and interpret the screening cascade

?Once we have a feasible TPP in mind, the next step is to begin the hunt for molecules to fit the bill.

At this point, model-based drug development will adjust the initial model from the TPP identification phase to ask a new question: If I want to discover a drug with this profile, what properties are most important to screen for? This involves running ‘sweeps’ for each of the parameters in the model, to understand which of the parameters have the most impact on outcome.

As a simple example, it might be imperative to first screen for soluble molecules. This insight alone can simplify the screening cascade by eliminating other candidate molecules earlier in the process. Modeling can also help you understand when to stop improving a property- how soluble is “soluble enough”? Often, project teams do this in real time, using subjective judgment. The problem is, it’s really difficult to do the mental math in our heads about the impact of a 20% increase in solubility- sometimes it could lead to a better drug, and sometimes it could end up making no difference. A PK/PD or Quantitative Systems Pharmacology model can help you take the guesswork out of that question, by providing you with estimates during screening.

So, the big advantage of designing and interpreting a screening cascade using mathematical models is that you can home in on the high-potential molecules after fewer experiments. That means saving time and money.

Step 3: Select the “right” molecule for pre-clinical development

?As the process moves closer to clinical trials, an important pivot occurs. The initial research, which usually involves mice, helps to assess the clinical efficacy of a molecule. (Does the potential drug shrink tumors, for example?) Animal experiments also provide information about toxicity. (How much of this drug is safe to put in the body?) But we must remember that the results apply specifically to the species tested.

To state the obvious, mice are not tiny humans. Although mice may make for good test subjects, there is more involved in designing a dosing schedule, for instance, than adjusting for the weight difference between Jerry the Mouse and your grandmother. When the lead molecule is selected based on what looks best in mice, the lead molecule will be the one that works best in mice. It takes careful model-based projection to understand what the therapeutic potential of a given molecule will be in humans.

Model-based drug development can account for interspecies differences in PK and PD. In other words, it can help answer questions about how a particular molecule will behave in the human body. How fast will it degrade? Where is the line between an efficacious and a toxic dose? And importantly, will a safe dose “work”—i.e., can a human being tolerate enough of this substance to achieve the hoped-for clinical effect (or does it have a therapeutic window)?

Modeling to project the therapeutic window in humans at the stage of lead selection is a good idea, because it allows you to focus development effort on the molecule that is most likely to succeed in the clinic (which isn’t always the molecule that had the flashiest result in your in vivo model of choice).

Step 4: Put together the Investigational New Drug application and clinical trial design

The next stage of development—the gateway to clinical trials—is the Investigational New Drug (IND) application to the FDA, which includes details regarding the clinical trial design.

Critical decisions loom here. For example, what dose will be first trialed in humans? And at what cadence will dosing increase in the search for the sweet spot where tumors shrink a lot but the patient lives and feels pretty good?

At first blush, such trial design choices may seem simple. One could proceed from an extremely low and assuredly safe dose and increase in regular increments to determine the maximum dose patients tolerate. Alternatively, dosing might increase at a sharp cadence until signs of negative side effects emerge—maybe patients start complaining of headaches at 50 mg—and then move more slowly from there.

The problem is that without a basis on which to judge cadence in advance, the decision will be less than ideal. In an information vacuum, researchers may be too conservative, starting at very low doses and increasing in extremely small steps, which results in a longer and more expensive clinical trial. In other cases, the plan may prove too aggressive at the point where toxicity impacts show up, putting patients in more danger than is prudent.

A better alternative is to base these choices on models capable of forecasting where side effects are most likely in humans so that a step plan can be more accurately calibrated. One can then escalate doses with confidence, maximizing trial efficiency and safety, with extra caution applied closely in advance of expected tolerance thresholds. These types of approaches combine population pharmacokinetic modeling (where each patient’s PK is modeled) with Bayesian statistics to provide an efficient way of estimating critical trial parameters even when data is very limited. Modeling is therefore valuable in selecting the starting dose (especially for situations where the traditional toxicological methodologies are not appropriate) as well as the dose escalation strategy.

Modeling is useful for other design choices for Phase I trials as well- ?as mentioned earlier, dosing schedule selection is fundamentally a modeling question, because the PK differs between preclinical species and humans. Biomarker sampling times?and interpretation are strengthened by modeling as well- a two hour PD sample in rats may behave very differently from the same sample in humans! Combination starting doses and escalation strategies are another area where modeling can help inform the strategy.

In each of these cases, it’s always possible to come up with answers without using modeling. However, model-based approaches are usually (much) cheaper and faster, and are more rigorous. They also have the benefit of speaking the language of the Agency- in our experience, over numerous INDs, the FDA review process goes more smoothly when modeling is involved.

Step 5: Execute on an efficient clinical development program

Finally, model-based drug development approaches are vital during the execution of a clinical trial. During Phase I, estimating the dose response relationships for PK, PD, efficacy and toxicity can provide crucial guidance. To make this happen with a model-based approach, patient clinical PK is analyzed using population PK models, and PK/PD modeling is done during real time. Bayesian statistical approaches can be used to provide a PK/PD relationship from the first dose- it starts out as a translational PK/PD relationship, and matures into a clinical PK/PD relationship as more clinical data is gathered. This can provide the clinician with real-time guidance about the dose ranges where biological activity and toxicity can be expected. Modeling adverse events using approaches such as categorical logistic regression can provide estimates for the dose-response relationship for toxicity.

When it comes to drug efficacy, many of the traditional metrics (such as RECIST criteria in cancer) are outdated, and discard much of the data that is gathered during a trial. Model-based approaches, such as disease progress models use all of the available data to make the clinical readouts less ‘noisy’, allowing for better decision-making. (Ping me on LinkedIn if you're curious to know more about this!)

Often during clinical trials, project teams will gather copious amounts of data, but can find it challenging to merge all of the different data points together to draw insights. Modeling the dose/ PK/ PD and dose/ PK/ toxicity relationships allows for rigorous integration of all the available data, and can squeeze more ‘signal’ out of the same dataset. ?

Model-based approaches can be used to create a ‘dashboard’ for the project team to get a sense of what to expect from the next dose escalation, which is vital for ensuring a rapid but safe dose escalation trajectory.

Getting started with applying model-based approaches to your program

How could you get started with model-based drug development? The on-ramp will vary based on the current stage of a research program. Modeling can add value at any stage along the discovery and development cascade- the earlier you start, the greater the overall impact. That said, each step of the process is quicker and cheaper with model-based approaches, so modeling provides an ROI at every step along the way.

We engage with clients at every step along the cascade, and offer our model-based drug development services that integrate PK/PD modeling with pharmacology and translational input as needed. As drug developers ourselves, we view modeling as a means to an end. This means that the goal is to move the project forward as quickly as possible, not to build the most complex model that can be built!

Admittedly, it can be intimidating to consider such a new approach as the one we described here. When presented with a promising but unfamiliar option, more information is almost always a good thing. For that, I would encourage you to visit www.fractaltx.com where we have posted a variety of white papers that cover the topics described in this post. You can request the white papers directly, or reach out to me via LinkedIn. If this sounds like it could be useful to you, let’s talk!