Date: October 11, 2026, 9:00 – 17:00

Location: Gaylord National Harbor, Maryland, USA

Instructor: Niklas Korsbo, PhD

Register Now!

Overview

“AI in pharmacometrics” nowadays refers to two almost distinct things: LLM and foundation model assisted workflows, and machine learning augmented analytics of our patient data. Both of these areas are currently rapidly changing what we can do in pharmacometrics. The LLMs quite self-explanatorily by their rapid improvement in performing complex tasks and the ML analytic approaches by deep and easily exploitable connections of NLME models to scientific as well as generative machine learning. Two parallel disruptions that both promise dramatic improvements to pharmacometrics. Improvements, but nevertheless disruptions that carry uncertainties about how to navigate the shift as an organization so as to neither fall behind nor overinvest in the wrong things.

In this workshop, we will take a zoomed-out view of where we are in these two AI/ML paradigm shifts and offer well-grounded speculations and discussions about where we are headed. That grounding is important and, in the workshop, we will mix current experiences and mathematical foundations that underlie our assessment of what the future holds with predictions and speculations about what capabilities will be unlocked and how bottlenecks and pain-points in the drug development pipeline will shift.

This is a conceptual and strategic workshop, not a coding session — no laptop is required and no machine learning background is assumed. The day mixes instructor-led grounding with lively discussion and attendee participation. The people in the room will collectively hold experiences and perspectives that no single instructor does, and we intend to draw on them.

LLM-based workflow automation

Current LLMs are already largely capable of carrying out much of the procedural work of a pharmacometric analysis. The bottleneck is no longer the capability of the model but the structure it operates in: codified analysis procedures, disciplined access to data and information, and enforced habits of double-checking and documentation. Building that structure is less like AI research and more like writing the training material for a rigorous new employee — it takes as much pharmacometric judgment as it takes an understanding of how LLMs operate. That carries an organizational consequence worth dwelling on: the work of codifying our workflows draws on the same people and hours as any other modeling investment we might make. We will discuss what this looks like in practice, what a well-instructed agentic workflow can and cannot be trusted with, and what this path does — and does not — do for the scientific frontier of the field.

Machine learning augmented pharmacometric analytics

Machine learning methods routinely scale to multi-endpoint, multi-modal data — including longitudinal imaging — in ways that classical NLME currently cannot. But this gap is not fundamental. The same mathematical machinery that makes ML flexible at scale can be integrated into the mechanistic, mixed-effects framework that pharmacometrics depends on. This makes it possible to use much richer patient information to assess how their primary endpoints change over time and what patient features associate with, predict, or modulate their outcomes. We will go through some of the opportunities here, like analyzing images over time to find rich multifactorial surrogate endpoints, utilizing multiple longitudinal biomarkers to more quickly inform predictions of the primary endpoint, as well as efficient data-driven identification of poorly understood structural NLME model components for increased model performance and workflow efficiency. We will assess the technical and scientific roadmap for these concepts to mature into greater real-world impact, consider the consequences of their maturation for organizational capabilities and shifts, and highlight both the potential benefits and risks with investments along this line of capabilities.

Further out, this same flexibility opens the possibility of larger platform models that bridge trials — models that accumulate what a portfolio of programs has already taught us so that each new program starts from knowledge rather than from scratch. If that materializes, it changes more than the analytics: a pharmacometrics function that learns across programs and feeds insight back into candidate selection, trial design, and portfolio decisions plays a rather different role in the pharmaceutical industry than one that de-risks a single trial at a time.

Learning Objectives

After this workshop, participants will be able to:

• Articulate how neural differential equations and latent variable models relate to classical NLME — technically, not just conceptually
• Assess where hybrid strategies (pretrained components, neural ODE backbones, fine-tuning across programs) apply in their own pipelines
• Identify which data practices and infrastructure investments pay off under different modeling strategies
• Make better-informed build/buy/wait decisions about ML capability in their organizations

Bring Your Team, Save More! Groups of 3+ qualify for a group discount. Email sales@pumas.ai to inquire.

Register Now!