Where PBPK Creates the Most Value

PBPK modeling has moved well beyond its historical role as a niche drug-drug interaction (DDI) risk assessment tool. Today, many R&D organizations are using PBPK earlier in development to inform candidate selection, formulation strategy, and clinical trial design. Put simply, PBPK is the strongest choice when development decisions depend on understanding how physiology, dose route, formulation, co-administration or patient variability influence drug exposure.
The science is evolving, and so are the ways in which teams operationalize it.

Common Misconceptions

One of the biggest mistakes is treating PBPK as a “plug-and-play” prediction engine, as critical thinking should always be used when building and applying PBPK models. Mechanistic models should be used to explore biological understanding, and the impact of the availability and quality of input data should be probed through parameter sensitivity analysis and a model verification strategy.
Another common misconception is that PBPK can answer every translational question. PBPK is exceptionally strong for exposure-related questions—absorption, distribution, DDI risk, organ impairment, pediatrics—but on its own, it is not designed to capture complex disease biology or systems-level pharmacology. That’s often where quantitative systems pharmacology (QSP) or PBPK combined with QSP becomes the better fit.
PBPK focuses on what is happening mechanistically, rather than on confirmatory clinical interpretation. If the question is what do the observed clinical data tell us, then clinical pharmacokinetics is a better option to quantify variability, efficacy, safety, or dose-response relationships observed in subjects.

Practical Examples

There are many common applications for PBPK modeling. It is now routinely used during drug discovery for in vitro to in vivo translation and pharmacokinetic modeling in preclinical species, which leads to the application of verified models for first-in-human (FIH) PK predictions. A recent publication provided a strategy for FIH PBPK modeling with real-world examples for challenging compounds.

PBPK is also used to predict the PK in special populations, as these subjects often experience major physiological changes that can significantly alter drug exposure, potentially affecting both efficacy and safety. For example, predicting differences in PK in pregnant subjects and fetal exposure for metabolically cleared compounds.

Another new publication detailed how one large pharma company is now employing high-throughput PBPK modeling for rodent pharmacokinetic (PK) predictions based on only machine learning predictions for input parameters. This approach facilitates lead series prioritization and streamlines in vivo PK and PK/PD testing during lead optimization, thus accelerating drug discovery timelines.

PBPK modeling can also be used, in some cases, to avoid DDI clinical studies or justify wider dissolution specifications, accelerating the program timeline and reducing costs.

Trends We’re Seeing

1. 1. PBPK is shifting earlier in the pipeline

Organizations are increasingly applying PBPK during lead optimization and preclinical translation, rather than waiting until the clinic. Early mechanistic insight can reduce downstream uncertainty around exposure, variability, and formulation performance.

2. Cross-functional decision-making is becoming more model-informed

PBPK is increasingly being used as a cross-functional decision-support tool across clinical pharmacology, formulation development, translational sciences, and regulatory strategy as part of broader model-informed drug development (MIDD) initiatives. For example, at candidate selection a company is likely to have a PBPK model for FIH predictions with the systemic disposition being preclinically verified by DMPK scientists, the absorption modeling being characterized by biopharmaceutical scientists, and the target patient population being assessed by clinical pharmacologists.

3. Greater emphasis on platform confidence, not just model fit

The quote “with great power comes great responsibility” is attributed to Spider-Man, but it now applies to PBPK as it becomes employed more widely and for higher levels of regulatory impact. This means regulators and internal governance teams are asking tougher questions around verification, assumptions, and reproducibility. The conversation is evolving from “Does the model fit observed data?” to “Can we trust this methodology prospectively to make key decisions?”

The Bigger Picture

PBPK delivers the most value when it is used as a decision-support framework — not simply a regulatory deliverable. The organizations getting the strongest return are the ones embedding mechanistic modeling into broader R&D strategy and governance, rather than treating it as an isolated modeling exercise.

If you are considering PBPK modeling for a program or want to understand where it fits within your overall modeling strategy, our team would be glad to discuss your specific questions.

Discuss My Project Needs with an Expert

• Autoimmune treatment remains inefficient: in search of sustained response, patients may spend years if not decades cycling through the treatment options available for their disease-label
• While disease-label-based assessments are conventional, the underlying immune processes can be viewed as a shared mechanistic system, under which different perturbations can result in autoimmune-driven symptoms, with substantial symptom overlap across those disease-labels
• This presentation describes a disease-label-agnostic framework combining mechanistic quantitative systems pharmacology (QSP) modeling with interpretable machine learning to derive biomarker-based precision medicine treatment-assignment rules
• Describe how a disease-label-agnostic autoimmune QSP framework can be combined with interpretable machine learning to guide precision medicine therapy-assignment
• Describe applications of the framework
• Describe benefits of the paradigm shift in thinking that the application suggests

By Noah Brostoff

AAPS PharmSci 360 2026, October 25-28, 2026, New Orleans, Louisiana

• Oral absorption-mediated drug-drug interactions (DDIs) are important in drug development. PBPK modeling provides a mechanistic approach to predict interactions resulting from changes in gastric pH, altered gastric and intestinal transit, and food effects. This presentation highlights recent PBPK applications, including pH-dependent DDIs, transit changes related to GLP-1 receptor agonists, and food effects. Case studies will demonstrate regulatory applications, and we will discuss emerging AI approaches.
• Upon completion, participants will be able to understand what absorption-related DDIs are and their potential impact on absorption and PK of orally administered drugs.
• Understand how PBPK modeling can help to assess absorption-mediated DDIs, such as pH-dependent interactions, food effects, and gastric emptying prolongation due to GLP-1 agonists.
• Understand the role of PBPK M&S in informing study design and regulatory decisions.

By Grazyna (Grace) Fraczkiewicz

AAPS PharmSci 360 2026, October 25-28, 2026, New Orleans, Louisiana

• Quantitative systems toxicology (QST) is a proven method for translating in vitro inputs to toxicity predictions
• Biologicsandothernon-small-molecule treatment modalities present different translational challenges
• QST can help translate novel NAM-derived data into toxicity predictions
• Understand how QST makes predictions from in vitro data inputs
• Understand the unique QST challenges presented by large molecules
• Understand how large molecule NAM-based in vitro data can be used as QST inputs

By Jeff Woodhead

AAPS PharmSci 360 2026, October 25-28, 2026, New Orleans, Louisiana

The Current Approach:

• Dissolution rate of ionizable compounds is historically treated in a simplified manner based on the solubility of the drug in bulk intestinal fluid at a static pH. Modeling dissolution provides important insight across multiple applications:
  • Formulation Optimization
  • Effect of Proton Pump Inhibitors (PPI)
  • Cross species predictions (dog vs. human)
  • Food Effect Modeling
  • Drug-Drug Interactions
  • Virtual Bioequivalence

The Problem:

• Ionizable drugs and/or excipients can change the pH at their own surface as they dissolve in weak intestinal bicarbonate buffer and gastric fluid leading to inaccurate dissolution predictions when assuming bulk pH.

The Solution:

• Improve existing chemical equilibrium approaches in literature to dynamically calculate the surface pH based on a drugs pKa(s) and important gastrointestinal fluid species.

By James Mullin, Deanna Mudie, Viera Lukacova

AAPS PharmSci 360 2026, October 25-28, 2026, New Orleans, Louisiana

When you ask the Monolix AI agent to perform the same task, it reads the dataset, asks a specialized subagent to write the corresponding mlxtran models, creates three Monolix projects, runs the estimation tasks, retrieves the results, and compares the models’ BICc in a single table.

What previously required a long sequence of manual operations is completed using one prompt.

The time saved is obvious, and similar examples exist across MIDD workflows, but it is not the whole picture.

What it means to trust an AI agent

We all see how fast LLMs are evolving and how much they can help us accomplish. We have also learned not to blindly trust the results they produce. Before an AI-assisted process becomes part of your work, you need to know it is at least as reliable, traceable and controllable as the manual steps it replaces.

In practice, that requires an environment in which you can delegate work, verify what was done, intervene when necessary, and keep the scientific decisions under your control. The rest of this article describes what such an environment has to provide, and how we built Composer to provide it.

Four questions worth asking of any AI-assisted workflow

Before you use AI in your MIDD workflows, there are four questions you can ask to evaluate it:

• Where do the numerical results come from?
• Can I inspect what AI agent did?
• Can I control what AI agent is allowed to do?
• What happens when something goes wrong?

The answers to these four questions determine whether an AI-assisted workflow can be relied on.

1. The numerical results come from the scientific engine, not from AI

The first question points to the most important feature.

Let’s consider the Monolix AI agent for PKPD modeling. It has access to a set of built-in tools to read and inspect data, generate mlxtran model files, set up a Monolix project, run parameter estimation and produce plots. The large language model (LLM) reasons about the task and decides which of these actions to take.

But when it requires a pharmacometric calculation, for example the SAEM parameter estimation, the LLM does not produce the numerical result itself.

The Model Context Protocol (MCP) interface connects the agent to MonolixSuite engine. It translates actions the agent requests into MonolixSuite commands, asks the software to perform calculations, and returns results to the agent. So when the agent needs parameter estimates or BIC values, the sequence is always the same:  the agent defines the action, MCP sends the corresponding command, MonolixSuite performs the calculation, and the agent receives the result.

A reliable workflow must never let an LLM invent numerical results that will end up in an analysis.

In the above example, the estimation is delegated to the scientific engine designed for it. You can open and run estimation in the same Monolix project the agent used, independently of the AI workflow, and obtain the same result. Numerical computations remain fully reproducible.

AI changes how the analysis is orchestrated, but it does not replace the engine that performs the science.

The same principle applies to structural model generation. When the agent needs an mlxtran model, a specialized subagent writes it from scratch based on your description, by translating a model from another software language, or by converting mathematical equations into the mlxtran syntax. The generated model is then validated against Monolix to confirm that the syntax is correct, that the model can be loaded and executed, and it is saved in the workspace as a file you can (and should) review before running the analysis.

The mlxtran subagent does generate model code, but that code is tested against the scientific engine and is stored as a reviewable file. You are never asked to trust a model definition you cannot verify.

2. You can inspect the workflow, not just its final answer

Reliable scientific work must be traceable: you should be able to see how a result was produced, not only that it was produced.

In Composer, you can follow the agent’s conversation live as it’s happening. Instead of the single black-box response typical of an AI chat, you see each action the agent takes, you can inspect the arguments passed to each tool call, and, whenever you want more oversight, you can approve tool calls before they are executed. The agent also provides datasets, R scripts, mlxtran model files, Monolix projects, tables and reports, which are stored in the Composer workspace as part of the session.

Consider the three-model example again. If the agent reports that the power model has the lowest BIC, you do not have to accept that as an isolated statement. Instead, you can open the corresponding Monolix project, inspect the model definition, estimation settings, convergence, parameter estimates and diagnostics, and compare the reported BIC against the value in the project itself.

Projects and model files the agent builds are standard MonolixSuite artifacts that exist independently of the AI session. That means the analysis can be reproduced later by someone who was not present when it was run, and who does not have the same tools you have, whether that’s a colleague inheriting the project or a regulatory reviewer reconstructing the work from the delivered files.

Reproducibility exists without the agent, without the chat session, and without you. This is also the difference between asking an LLM a scientific question and using AI agents to operate scientific workflows: the answer stays connected to the calculations, files and actions that produced it.

3. You decide how much you delegate

Not every AI task must be supervised in the same way.

Some tasks are already defined in the analysis plan and reproduce a known sequence of steps, such as building three predefined models or running an established analysis. In cases like these, you may want to let the agent execute several operations and review the result at the end.

In exploratory work, you may ask the agent to inspect diagnostics, investigate an unexpected pattern, formulate an alternative hypothesis or prepare a further model for evaluation, all under close supervision.

The required level of supervision depends on the scientific task.

Composer lets you set these constraints in the prompt or in the agent instructions. Individual tool calls can be reviewed before they run, and workflows can be executed either more autonomously or with frequent checkpoints.

This is also where a common and reasonable question is usually asked: if the agent’s behaviour is not fully deterministic, how can the workflow be controlled? It is true that the same prompt will not reproduce the same chat text word for word, and the agent may take a slightly different path through the tools. What matters is that this non-determinism is limited, because you can stop the agent at any point, require it to seek your approval before a given step, and specify in the prompt the exact steps it must complete. You do not need the orchestration to be identical every time when each thing the agent does is visible and verifiable.

The important principle is that you decide what to delegate to an agent.

Using an agent does not transfer responsibility for the analysis. You still define the scientific objective, provide constraints, review the decisions that matter and decide whether the evidence supports a conclusion. The agent acts as a co-scientist: it executes work, gathers information, and proposes next steps, while you guide the analysis.

4. Reliable AI workflows must assume errors will happen

AI agents will sometimes be wrong.

A generated model may contain invalid syntax, a dataset may not have the expected structure, an estimation may fail to converge, and an agent may occasionally choose a next step that is technically valid but scientifically inappropriate.

A reliable system must be built on the assumption that errors will appear, and they need to be visible.

In Composer, when a tool or MonolixSuite returns an error or a warning, that information is passed back to the agent and displayed in the chat. In many cases the agent can diagnose an execution problem and correct its own action.

But not every problem is technical. When an estimation does not converge the agent can inspect the result and suggest possible next steps, but deciding whether the cause is initialization, model misspecification, insufficient information in the data or something else is a scientific judgment.

What a reliable AI-assisted workflow should give you

The standard you should require from an AI-assisted workflow comes back to the answers to four key questions.

Where did the numerical results come from?
In Composer, they come from the validated scientific engine, not from AI.

Can I inspect how they were obtained?
Yes. Tool calls, arguments, generated files and projects, and the conversation are available in the Composer project environment.

Can I control what the agent is allowed to do?
Yes. Composer allows you to adapt the level of autonomy and review each task.

What happens when something goes wrong?
You see every error, warning, and decision the agent makes. The final judgment stays with you.

Composer changes how much of the execution has to be done manually, so more of your time can be spent on analysis and critical decision-making instead of the mechanics of running the workflow.

But it does not change that validated engines perform the computations, and that you guide the workflow, review the evidence and decide what the results mean.

You do not need to blindly trust an AI agent to use it, but you do need an environment in which you can verify what was done, intervene when necessary and keep the scientific decisions under your control.

Schedule a demo if you’d like to see this for yourself.

Request a Demo

• Ziftomenib is a potent, selective, oral menin inhibitor that is U.S. FDA-approved as monotherapy at 600mg QD for adults with relapsed/refractory (R/R) NPM1-m AML and no other satisfactory treatments.
• It has shown clinical activity and tolerability in combination with venetoclax/azacitidine (ven/aza) and cytarabine/daunorubicin (7+3) standard-ofcare (SOC) chemotherapies in adults with newly diagnosed (ND) or R/R, NPM1-m or KMT2A-r AML
• KOMET-007 (NCT05735184) is an ongoing, multi-cohort, open-label, Phase (Ph) 1a/b (dose-escalation and -expansion) study of ziftomenib combined with SOC therapies, including ven/aza and 7+3, in adults with NPM1-m or KMT2A-r AML
• A QSP model of AML progression was developed and used to characterize the efficacy and safety of these ziftomenib combinations, to support further development

By Cameron Meaney, Ashley Markazi, Ryan Suderman, Mollie Leoni, Amitava Mitra

ACoP 2026, October 11-14, 2026, National Harbor, Maryland

• With the introduction of the new approach methodologies (NAMs) initiative, there is a focus on finding alternatives to animal studies to validate a drug’s safety before administering in humans.
• Quantitative systems toxicology (QST) models are uniquely positioned to address this need. However, these models often rely on inputs from in vitro data with the ideal not always available.
• In this work, kidney toxicity predictions were made using information gathered in a variety of cell lines (kidney and others).
• The objective of this proof of concept was to find ways to strengthen the predictive ability of QST models using less-than-ideal in vitro inputs.

By Celeste Vallejo and Jeffrey L. Woodhead

ACoP 2026, October 11-14, 2026, National Harbor, Maryland

By Xinyuan (Susie) Zhang
ACoP 2026, October 11-14, 2026, National Harbor, Maryland