So the question on your mind is not whether agentic AI is coming. It is this: can you stake a regulatory submission on it?

That is the question this piece is about, because it is the one that matters most in our industry. A drug program is a chain of decisions that stretches over years and absorbs hundreds of millions of dollars. Every link in that chain has to be defensible—to a regulator, to an internal quality team, to the next scientist who picks up the model two years from now and needs to understand exactly how a conclusion was reached. An AI agent that produces a brilliant insight you cannot reproduce, cannot audit, and cannot trace back to a validated computation becomes a liability in that world, however impressive the insight.

The questions that separate trustworthy agentic AI from the rest

If you are evaluating agentic AI for regulated science, four questions tell you most of what you need to know:

• What actually produces the result—a language model, or a validated scientific engine?
• Can the result be reproduced—exactly, every time, from the same inputs?
• Can every decision be traced and replayed—months or years later, by someone who wasn’t there?
• Does human expertise stay accountable—with the scientist directing the work, not deferring to it?

These are the questions we built Composer, our AI-native platform for model-informed drug development, to answer. At its heart is an AI co-scientist: an agent that reasons about the scientific problem and orchestrates our validated engines through natural language, while the scientist stays in control. Here is how.

Our thesis: the insight is only as trustworthy as the engine beneath it

The conviction at the center of how we build, and the thing our customers respond to most strongly: an AI-generated insight is only as trustworthy as the validated engine that produced it.

A large language model can summarize a mechanism, propose a hypothesis, even sketch a model structure. What it cannot do—and should never be asked to do—is be the science. The numerical simulation, the parameter estimation, the mechanistic ODE solver that a drug decision rests on must come from an engine that has been validated against reference models and benchmark datasets, that produces the same output every time it is run, and that can be audited independently of any AI in the loop. A deterministic, validated engine is a world away from a chatbot dressed up as science.

So in the Composer ecosystem, the AI orchestrates the engine rather than replacing it. The co-scientist is designed to reason about the problem and decide what to do; the validated engine does the computing. We keep a clean line between the reasoning layer, which is probabilistic and fast-moving, and the computation layer, which is deterministic and qualifiable. The AI is intended to be an enhancement—a powerful one—but it is deliberately kept off the regulatory critical path. Every engine in Composer remains fully usable, and fully validatable, without any AI at all.

That architecture is what lets us answer the second and third questions directly.

Reproducibility. Run the same analysis with the same inputs and you get the same result—because the work is done by deterministic engines, not by a model that improvises. When a scientist is ready to move from open-ended exploration to a standardized process, the co-scientist can capture the entire interaction as a deterministic Workflow: an inspectable, versionable artifact that runs the same way every time, with no AI in the execution path. Exploration becomes production without losing rigor.

Replay-ability. Every analysis leaves a complete, traceable record—which engine version ran, with which parameters, producing which result, leading to which decision. A colleague, a regulator, or the same scientist two years later can replay the reasoning end to end. The trail is the explanation. In a field where “right for the wrong reasons” is a genuine scientific failure mode, the ability to show your work is part of the science itself, well beyond any compliance checkbox.

For three decades, our engines—GastroPlus®, MonolixSuite™, ADMET Predictor®, and our mechanistic QSP platforms—have carried exactly this burden of proof inside regulatory filings around the world. They are the foundation the agents stand on.

Why the BioNeMo Agent Toolkit is foundational to this

A validated foundation answers the trust question. But the agents that orchestrate that foundation also need to be genuinely fluent in biology, not merely fluent in language. This is exactly where the NVIDIA BioNeMo Agent Toolkit becomes foundational to our strategy.

General-purpose reasoning gets an agent surprisingly far. But drug development runs on domain-specific understanding: the structure of a protein, the pharmacology of a target, the patterns buried in genomics and clinical data, the mechanistic logic of a disease. The BioNeMo Agent Toolkit—NVIDIA’s life-sciences agent stack, including Nemotron models, NemoClaw blueprints, and OpenShell secure runtime—provides domain-specific models, NVIDIA NIM microservices, libraries, and frameworks built for life sciences. It is what lets an agent act on biological data rather than merely describe it, and it is what bridges general-purpose AI reasoning with real-world biological computation.

This is the combination that matters: our validated engines seek to bring scientific truth that holds up under scrutiny; the BioNeMo Agent Toolkit brings biological breadth and acceleration. Neither is sufficient alone, and the scientist directs both.

How we are using the BioNeMo Agent Toolkit today—and where we are going

The collaboration is already underway across our ecosystem on three fronts.

Grounded in evidence. Our agents are incorporating NVIDIA Nemotron Parse for scientific literature parsing and extraction—pulling parameters, mechanistic relationships, and quantitative data out of the dense PDFs, tables, and figures that hold the field’s accumulated knowledge. Extraction quality is our goal: a misread table becomes a bad parameter becomes a flawed model. By grounding our agents’ knowledge bases in high-fidelity extraction—with provenance back to the source—we let the co-scientist reason over what the literature actually says and let the scientist verify the evidence behind any recommendation. This is how an agent earns the right to propose a parameter rather than hallucinate one.

Accelerated by computation. Together with NVIDIA, we are collaborating on nvQSP to bring CUDA-optimized ODE solvers to our mechanistic modeling engines—GPU-accelerated QSP simulation that reduces the time to evaluate virtual populations and explore competing hypotheses. The science of a QSP or PK/PD model lives in systems of differential equations, and solving them is the computational bottleneck. GPU-accelerated solvers return the same answer the CPU would, while changing what becomes possible: larger virtual populations, broader parameter-space exploration, agentic search across many candidate models in the time it used to take to run one. Faster computation does not change the science; it expands the number of scientific questions a team can afford to ask.

Expanded by specialized biological models. Our vision goes further, and we are building toward it deliberately. We intend to host NVIDIA’s frontier biological models directly within the Composer ecosystem and to power a new generation of Composer Agents on BioNeMo Agent Toolkit infrastructure—new classes of biological reasoning available to the co-scientist as first-class, orchestratable tools, sitting right alongside our validated simulation engines. Every one of those new capabilities inherits the same discipline as the rest of the platform: scoped, version-traceable, and validated for what it is claimed to do, with the AI layer kept off the critical path. Biological frontier models and decades-validated engines, in one ecosystem, under one standard of trust.

What this means for scientists and for discovery

Strip away the architecture and the partnership mechanics, and here is what changes for the person doing the work — which brings us to the fourth question, the one about human accountability.

Agentic AI removes friction from the process, not the scientist from the science. Hypotheses get generated and tested faster, because the distance between “I have an idea” and “I have a result” collapses from weeks to hours. Experimental scientists who were never going to become expert modelers can now direct sophisticated modeling through natural language — the expertise is in the platform, available to them. And the collaboration between human scientist and AI agent becomes a real partnership rather than a novelty, because every recommendation remains inspectable, reproducible, and subject to human review. Scientific accountability stays exactly where it belongs: with the experts directing the work.

That is the future we are building toward: agentic science that moves at the speed of AI and holds up to the scrutiny of a regulator, in the same breath.

Agentic drug development needs a foundation scientists can trust. We have spent thirty years building the validated engines that scientific truth rests on; NVIDIA BioNeMo Agent Toolkit extends its reach into new domains of biology; and the scientist stays firmly at the controls of both. That is the partnership—and the standard—we are bringing to BIO 2026.

By Erik Guffrey, Co-Chief Product & Technology Officer

During model structure selection, to ensure biological and clinical plausibility

Interpretability helps distinguish between models that merely fit the data and those that meaningfully represent underlying physiological or pharmacological processes. A structurally interpretable model allows reviewers to understand why the model works, not just that it does work, reducing the risk of overfitting or spurious relationships.

When evaluating covariates, to distinguish meaningful relationships from statistical artifacts

Covariate inclusion should be driven by biological plausibility and clinical relevance, rather than statistical significance alone. Interpretability enables assessment of whether observed relationships are consistent with known mechanisms or are likely artifacts of the dataset. This is particularly important when covariates influence dosing recommendations or subgroup labeling.

During simulation and extrapolation to new populations or dosing regimens

When models are used beyond the observed data, such as in pediatric extrapolation, special populations, or alternative dosing strategies, interpretability becomes critical for assessing credibility. Mechanistic understanding supports confidence that the model will behave appropriately under new conditions and that predictions are not driven by unsupported assumptions.

When communicating results to regulators to ensure transparency and credibility

Interpretability plays a central role in regulatory communications. Reviewers must be able to follow the logic from assumptions to conclusions, particularly for high-impact decisions. Clear, interpretable models facilitate efficient review, reduce ambiguity, and increase confidence that conclusions are robust and reproducible.

Interpretability is what allows a model to move from a technical artifact to regulatory evidence. Yet even interpretable models can lose credibility if the underlying process is not transparent. In a future post, we will examine the most common modeling mistakes that reduce reviewer trust and how to avoid them.

By Sandra Suarez-Sharp

This manuscript highlights key sessions from the In Vitro Release and Dissolution Testing (IVRDT) and Oral Biopharmaceutics and Absorption Modeling (OBAM) communities at the 2025 AAPS PharmSci 360 Annual Meeting (November 9–12, San Antonio, TX). Presentations emphasized a shift toward mechanistic and predictive in vitro and in silico tools, including enzymatic dissolution media, physics-based particle dissolution modeling, physiologically based biopharmaceutics modeling (PBBM) applications for food-effect assessment and specification setting, advanced gastrointestinal simulation platforms, AI-driven formulation optimization, and the complementary role of preclinical animal studies. Collectively, these sessions underscored the value of integrating biorelevant experimentation with computational modeling to streamline and de-risk oral drug development.

By Jozef Al Gousous, Pradnya Bapat, Catarina I. D. Chendo, Nikoletta Fotaki, Vivian A. Gray, Andre Hermans, Rob Tzuchi Ju, Aleksander Mendyk, Tahseen Mirza, Deanna Mudie, Ishai Nir, Connor O’Farrell, Sanjaykumar Patel, Niloufar Salehi, Divyen Shah, Jie Shen, Balint Sinko, David C. Sperry, Sandip Tiwari, Yasuhiro Tsume, David B. Turner, and Yanxing Wang

Drs. Zhu and Booth review some of the recent approaches and data collection to select doses, including randomized dose cohorts, before segueing into the next big hurdle in dose selection, that being dose selection of combination regimens. Questions about how to select the dose of new therapeutic added to another approved agent, to established regimens with multiple agents, and two novel therapeutics are discussed.

Model-informed drug development (MIDD) has been an invaluable set of tools, which is playing an increasingly important role for oncology dose selection, especially for novel modalities such as monoclonal antibodies, bispecific antibodies, antibody-drug-conjugates, oligonucleotides, or cells. MIDD tools have been invaluable for the exploration of the treatment of various cancers. These novel modalities are associated with unique mechanisms of actions and pharmacology features. It has been shown that modeling and simulation tools that are developed based on small molecules can be readily applied to support new drug development for these novel modalities. The focus of the modeling and simulation work may differ to reflect the unique issues in the development programs for these novel modalities. Drs. Zhu and Booth also share some of the modeling work that has been conducted for dose selection of novel modalities as case examples.

This blog post is the first installment of a series outlining relevant topics such as practical principles, common pitfalls, and review-driven insights that extend beyond formal guidelines, focusing on what drives reviewer confidence in a modeling approach. In this post, we’ll focus on model transparency.

What Model Transparency Means in Practice

Transparency enables reviewers to understand the model in the context of use, including its structure, assumptions, constraints, how it supports the submission, to independently verify results, and evaluate how these elements influence conclusions. In practice, this level of transparency is achieved through a set of core, interdependent elements that ensure traceability, reproducibility, and scientific rigor throughout the modeling workflow. Transparency is not a single action, but is built through multiple deliberate practices across the modeling lifecycle, including:

1. Explicit linkage between modeling objective and regulatory decision (i.e., context of use)
2. Clear description of model structure, assumptions, constrains, and covariate relationships with scientific justification
3. Traceable parameter values and datasets with documented sources and derivations
4. Executable model code and defined software environments
5. Scientifically justified assumptions and covariate relationships
6. Diagnostics aligned with conclusions
7. Clear communication of uncertainty and limitations

What Documentation Do Regulators Expect?

Regulators expect submissions to provide a complete, reproducible, and decision-focused package. While guidance documents outline requirements, review practice consistently emphasizes clarity, traceability, and reproducibility. To meet these expectations, sponsors must ensure that key components of the modeling workflow are clearly documented and aligned with the decision being supported, including:

1. Clear question of interest linked to the regulatory decision (e.g., dose selection, labeling)
2. Final analysis parameter values and/or datasets with full traceability and documented derivations
3. Model files, scripts, software versions, and instructions
4. Diagnostics package consistent with agency expectations
5. Justified handling of BLQ data, outliers, and missingness with sensitivity analyses
6. Clinically meaningful covariate evaluation
7. Simulations that directly support conclusions and reflect uncertainty
8. Explicit discussion of limitations and their impact on decisions

Transparency is not merely a matter of providing complete information; it is about clearly exposing the scientific logic and assumptions that underly a model. True transparency allows others to understand how conclusions are reached, to rigorously test those conclusions, and to judge their credibility with confidence. Without this level of visibility, even a well-documented model can fall short of being trustworthy.

In the next article, we turn to the context in which interpretability matters most. Specifically, where models play a direct role in shaping regulatory decisions. It is at these critical junctures that the need for clear, defensible, and fully interpretable modeling becomes paramount.

Background/Objectives: Propranolol (PROP) is a non-selective β-blocker widely prescribed for cardiovascular and neurological disorders. Its pharmacokinetics (PK) are highly variable, and co-administration with omeprazole (OME), a CYP2C19 substrate and inhibitor, may alter systemic exposure. Herein, this study aimed to investigate factors influencing PROP PK variability and evaluate the effect of OME coadministration using physiologically based pharmacokinetic (PBPK) modeling and population PK (popPK) analysis.

Methods: PBPK models for PROP and OME were developed and validated against published data. DDI simulations were conducted across clinically relevant dosing regimens. A two-period fixed-sequence virtual trial of 125 subjects was simulated with PROP alone and PROP combined with OME. Population PK (popPK) analysis was performed on simulated plasma concentration data to identify covariates affecting PROP disposition and quantify DDI magnitude.

Results: PBPK models were successfully developed and validated. PROP disposition was best described by a two-compartment model with linear elimination. Health status was found to influence clearance, and body surface area (BSA) affected the central volume of distribution. Co-administration with OME increased PROP exposure, with larger effects in patients with renal impairment. Simulated plasma concentrations remained below established toxicity thresholds.

Conclusions: Virtual clinical trials integrating PBPK and popPK modeling provide a robust approach to identifying key determinants of PK variability and DDI risk. Although these findings were not directly translated to clinical observations, this helps identify sources of PK variability in PROP treatment settings and factors that may intensify its interaction with OME, thereby supporting model-informed precision dosing to enhance safety and efficacy.

By Lara Marques and Nuno Vale

Simulations Plus, Inc. (Nasdaq: SLP) (“Simulations Plus” or the “Company”), a global leader in model-informed and AI-accelerated drug development that advances biopharma innovation, today announced that Shawn O’Connor, Chief Executive Officer, will be participating in the 23rd Annual Craig-Hallum Institutional Investor Conference taking place in Minneapolis, Minnesota. Mr. O’Connor will host one-on-one meetings with institutional investors on Thursday, May 28, 2026.

For more information about the events or questions about registration, interested parties should reach out to their contacts at Craig-Hallum.