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

QSP development has traditionally required scientists to make a series of choices about which biological mechanisms to include, how to structure the equations, and which parameters to optimize. These decisions are made with limited ability to test against plausible alternatives, but Turin is designed to change that. The engine enables systematic exploration and evaluation of candidate mechanistic structures against the available evidence in parallel, helping scientists identify the ones that best balance biological complexity, explanatory performance, and practical utility. Turin’s capabilities are further enhanced by the technical collaboration with NVIDIA that was announced in May 2026.

Under this collaboration, the pharmaceutical partner will use Turin within a hosted environment, supported by Simulations Plus scientists, who will provide scientific guidance throughout model evaluation. The approach combines AI-scale computational exploration with expert human scientific judgment, keeping the partner’s scientists directly engaged in every step of the QSP model-building and evaluation process. The partnership is intended to rapidly expand into more therapeutic development programs.

“QSP can provide powerful insight into complex biological systems, but the time and expertise required to build and evaluate these models has historically limited where and when they can be applied,” said Shawn O’Connor, Chief Executive Officer of Simulations Plus. “Turin is designed to break down that barrier and make QSP practical for a broader range of development decisions. Working alongside a leading pharmaceutical organization allows us to tailor this approach to meaningful challenges in a real-world development setting while continuing to advance the scientific capabilities of the engine.”

Organizations interested in learning more about Turin or exploring potential partnerships may contact Simulations Plus: https://www.simulations-plus.com/contact/

In this panel discussion, leaders from Pfizer, Bristol Myers Squibb, Incyte and Simulations Plus for a candid discussion of how organizations are approaching AI in MIDD: where it is already creating value, where the technology is evolving, and what it takes to integrate AI without compromising scientific or regulatory expectations.

Topics discussed include:

• Where AI is delivering measurable value in MIDD today
• How organizations are approaching build, buy, and partner decisions
• What is required to integrate AI responsibly into scientific and regulatory workflows
• How the role of scientist is evolving, and which skills will become increasingly important

Whether you are already implementing AI or still defining your strategy, this discussion will help you identify the scientific, organizational, and technology considerations that should shape your approach.

The transaction was originally announced on June 16, 2026. At the effective time of the merger each Simulations Plus shareholder received $18.50 in cash per share. As a result of the completion of the transaction, Simulations Plus has become a privately held company, and its common stock will no longer trade on the Nasdaq Global Select Market.

As previously announced, in connection with closing, Simulations Plus is being combined with Chemical Computing Group (“CCG”), an Altaris portfolio company and leading provider of molecular modeling and drug discovery software. Together, the organizations will create a broader platform spanning drug discovery, modeling, simulation, and development, enhancing their ability to serve customers across the pharmaceutical research and development lifecycle. Within the combined business, Simulations Plus will continue to operate under its own name and brand.

“Today marks an exciting new chapter for Simulations Plus,” said Shawn O’Connor, Chief Executive Officer of Simulations Plus. “Our mission has always been to help our customers accelerate the discovery, development, and commercialization of innovative therapies through advanced modeling, simulation, and AI-driven technologies. By joining CCG, we believe we are well positioned to accelerate investment in innovation, expand our capabilities, and expand the solution we provide to customers across the biopharmaceutical ecosystem. We want to thank our employees, customers, partners, and shareholders for their support throughout our journey as a public company.”

“We are delighted to welcome the Simulations Plus team to CCG,” said Paul Labute, Chief Executive Officer at CCG. “Simulations Plus has established itself as a trusted leader in biosimulation and model-informed drug development, supported by its science, differentiated technology, and deep customer relationships. We look forward to partnering with the company’s talented team to support continued growth and innovation while advancing solutions that help improve the efficiency and success of drug development.”

The combined organization will leverage the complementary technologies, scientific expertise, and customer relationships of both companies to accelerate innovation and provide an expanded suite of software and services supporting biopharmaceutical research and development.

When models are introduced early, they become tools for learning and planning, not just for confirmation. This allows teams to proactively shape development rather than react to data after the fact.

If you are still utilizing modeling at later stages of development, you may want to consider a shift—and in this blog post, we’ll explain why.

Key benefits of early integration

Leading pharma companies are expanding their use of modeling and simulation, and there are five main reasons.

1. Guiding data collection aligned with key decisions
   Models help define what data are needed to answer critical questions (e.g., dose selection, exposure-response), ensuring studies are designed to be decision-relevant rather than exploratory.
2. Enabling iterative model refinement as new data emerge
   As additional data become available, models can be continuously updated, improving their predictive performance and reducing uncertainty over time.
3. Identifying knowledge gaps early in development
   Early modeling highlights limitations and gaps in understanding (e.g., insufficient covariate coverage, lack of exposure range, uncertainty in target distribution and/or expression), allowing focused data collection before pivotal studies.
4. Reducing uncertainty before key decision points
   Through iterative model refinement, uncertainty can be more systematically characterized, quantified, and reduced ahead of high-impact decisions such as Phase III trial design, labeling strategies, and biowaiver determinations. This process enables more informed decision making by explicitly evaluating how assumptions, variability, and data limitations (e.g., fit-for-purpose data) influence predicted outcomes.
5. Strengthening overall program confidence and regulatory credibility
   A prospectively integrated modeling strategy demonstrates a systematic and scientifically grounded approach, which aligns with regulatory expectations for model-informed drug development.

Model risk and model impact in a regulatory context

Regulatory evaluation of models increasingly incorporates a risk-based framework that considers both model risk, the potential for incorrect or misleading decisions due to model limitations, and model impact, the extent to which the model influences a regulatory or clinical decision.

Model risk arises from multiple sources, including structural assumptions, data limitations, parameter uncertainty, and extrapolation beyond observed data. Importantly, risk is not only about whether a model is “correct,” but about the consequences of being wrong. A model used to explore hypotheses carries relatively low risk, whereas a model directly informing dosing, labeling, or waiving clinical studies carries substantially higher risk. Notably, these two are not mutually exclusive, where model exploration of hypotheses may be the only “direct” mode of addressing the impact of the hypotheses on critical decisions. In these more complex settings, the need for an integrated weight-of-evidence approach is paramount.

Model impact, in contrast, reflects how heavily . Models may play a supportive role (e.g., contextualizing results) or serve as primary evidence (e.g., dose justification, pediatric extrapolation, or biowaiver decisions). As model impact increases, so does the expectation for rigor.

Regulatory expectations scale with the combination of risk and impact:

• Low-impact / low-risk models may require standard documentation and basic evaluation.
• High-impact / high-risk models require more extensive documentation, including clear language interpreting the simulation results and the limitations of the interpretation, sensitivity analyses, and evaluation of alternative assumptions.

High-impact and high-risk models demand the highest level of transparency, reproducibility, and credibility, often including multiple lines of evidence to support conclusions.

In practice, this means that model evaluation is not one-size-fits-all. Instead, the level of validation, documentation, and scrutiny should be proportionate to how the model will be used and the potential consequences of incorrect decisions.

Case study: dose selection and model risk

A sponsor submitted a population PK/PD model to support dose reduction for an oncology therapy. The model was intended to directly inform labeling, positioning it as a high-impact application, where model outputs would serve as primary evidence for a regulatory decision.

During initial review, several concerns were identified. Dataset derivations were not fully traceable to source data, covariate inclusion lacked clear biological or clinical justification, and sensitivity analyses, particularly around key assumptions such as BLQ handling, were limited. While the model demonstrated reasonable statistical performance, these gaps reduced transparency and hindered reproducibility.

Given the model’s high impact, these limitations translated into elevated model risk. From a regulatory perspective, insufficient traceability and lack of robustness testing raised concerns about whether the proposed dose reduction was reliably supported and whether alternative assumptions could materially change the conclusions.

To address these issues, the sponsor revised the submission by strengthening both transparency and model evaluation in a manner aligned with the model’s impact:

• Providing fully traceable datasets

All analysis datasets were clearly linked to source data, with documented derivations enabling independent reconstruction.

• Enhancing reproducibility with executable workflows

Model code, scripts, and run instructions were standardized and fully executable, allowing reviewers to reproduce key analyses without ambiguity.

• Adding sensitivity analyses for key assumptions

Alternative approaches to BLQ handling, covariate inclusion, and structural assumptions were evaluated and reported to demonstrate the robustness of conclusions.

• Demonstrating robustness under alternative scenarios

Simulations were expanded to show that the proposed dosing strategy remained valid across plausible ranges of uncertainty. These enhancements allowed reviewers to independently run the analysis and assess the solidity of conclusions under different assumptions. As a result, perceived model risk was substantially reduced, and confidence in the model increased. Ultimately, the model was accepted as credible evidence to support the dosing strategy. Importantly, the underlying model did not fundamentally change; the key shift was in transparency, reproducibility, and alignment of evaluation with model impact.

Transparent modeling is essential for managing model risk and supporting high-impact regulatory decisions. When transparency, reproducibility, and scientific justification are aligned with model impact, confidence in model-informed decisions increases significantly. Organizations that integrate modeling early, document thoroughly, and align validation with risk are better positioned for efficient regulatory review and successful outcomes.

If you’d like support with your regulatory strategy, schedule a 15-minute call with our experts.

Schedule a Call

Background and Objective
Understanding interindividual pharmacokinetic variability and potential for drug interactions is of clinical importance for pomalidomide given its long-term use in patients with relapsed and/or refractory multiple myeloma (RRMM). This study aimed to develop a physiologically based pharmacokinetic (PBPK) model of pomalidomide that integrates multiple sources of pharmacokinetic (PK) variability, including RRMM-related physiology, ethnicity, hepatic impairment, food effects, drug interactions, and cigarette smoking.

Methods
A PBPK model was developed incorporating in silico-predicted and experimentally measured physicochemical and biopharmaceutical properties, together with physiological parameters representing disease state, liver function, prandial status, and ethnic differences. Changes in PK and systemic exposure associated with multiple sources of variability, evaluated as shifts in central tendency, were predicted by the model.

Results
The model provided reliable estimates of observed PK profiles across all evaluated scenarios, with more than 90% of predicted PK metrics falling within 0.80- to 1.25-fold of clinically reported values. Changes in pharmacokinetics associated with RRMM in people of European and Japanese ancestry, as well as severity-graded hepatic impairment, were reasonably predicted, despite slight overestimation of systemic exposure in severe hepatic impairment (prediction-to-reported ratio ≈ 1.55). Physiologically based pharmacokinetic model predictions of the effect of co-administration with high-fat, high-calorie meals (15–30% reduction in Cmax) were also consistent with clinical observations in both European and Chinese cohorts. In addition, the PBPK model demonstrated good predictive performance in simulating moderate interactions perpetrated by fluvoxamine (cytochrome P450 1A2 [CYP1A2] inhibition) and tobacco smoking (CYP1A2 induction) and the negligible effect of ketoconazole co-administration (CYP3A4 inhibition), with predicted-to-reported Cmax and AUC ratios all within the 0.90–1.05 range.

Conclusions
Overall, this work highlights the utility of a PBPK modelling framework in providing mechanistic insights into multiple sources of pomalidomide PK variability, with potential broader applicability to other drugs.

By Jeffry Adiwidjaja, Alan V. Boddy & Andrew J. McLachlan

AI can process vast amounts of information, identify patterns, and accelerate workflows at remarkable speed, but it doesn’t understand why any of that matters. Experienced researchers do. The future of drug development should not be viewed as AI replacing scientists. It should be viewed as scientists becoming more empowered through AI-enabled tools and technologies, using AI as a co-scientist to support better decisions.

As scientists, we must do what we’ve always done—build the future instead of letting it happen to us. We must harness AI and build processes that leverage its strengths and constrain its limitations. We must evolve our understanding of our own roles and how our human intelligence, curiosity, and creativity contribute to the research and development of new treatments.

(If you’re not sure where to begin, my colleague Priyata Kalra wrote a post that I highly recommend.)

Scientific progress has never simply been about generating outputs. It is about asking the right questions, understanding biological relevance, challenging assumptions, and interpreting results within the broader context of human health and disease. These are fundamentally human capabilities. AI may support decision-making, but scientists determine whether those decisions make sense. They understand the limitations of datasets, the complexity of biological systems, and the nuances that sit behind every model or simulation.

The rise of AI makes scientific leadership more important than ever, because the volume and speed of information generation are increasing dramatically.

Without experienced scientists guiding that process, there is a real risk of overconfidence in AI outputs that may not fully reflect biological or clinical reality. The industry must be careful not to rush towards a future of fully AI-developed drugs before the science, regulatory confidence, and real-world evidence are truly there to support it.

I was recently asked whether I thought AI would replace clinical trials over the next 10 years. My immediate response was simple: would you personally take a drug fully designed by AI? Without human insight and oversight guiding the process, I doubt many of us would say yes.

The organizations that achieve the greatest success with AI over the next few years will be those that keep scientists firmly at the center of innovation. Technology alone is never enough in this industry. What drives meaningful progress is the combination of scientific expertise, domain knowledge, regulatory understanding, and advanced computational capability working together.

As scientists, we should not feel threatened by AI. It has the potential to remove inefficiencies, accelerate insight, and support more informed decision-making. But the direction, interpretation, and responsibility must still come from people with deep scientific understanding.

As companies race to integrate AI into their workflows and build increasingly data-driven platforms, the organizations that create the most value will not simply be the ones adopting AI the fastest. They will be the ones ensuring their scientists are leading how it is applied, interpreted, and trusted.

If you’re interested in learning how your organization can leverage a combined AI and MIDD approach, schedule a call or join us for our upcoming webinar, AI in MIDD: From Hype to Practical Application.

Register Now