AI in Pharmacometrics: A Strategic Vision for the Next Five Years | MD, USA

Format: One-day workshop

Primary Audience
Pharmacometrics leaders, R&D decision-makers, and quantitative scientists navigating how machine learning fits into their modeling strategy. No laptop required — this is a conceptual and strategic workshop, not a coding session.

Overview
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. The practical question facing most teams is *where on the spectrum between mechanistic constraint and data-driven flexibility* their modeling strategy should sit, and what that choice implies for data practices, team structure, and program design.
This workshop builds a shared technical foundation in the morning — from classical NLME through Neural ODEs and deep NLME to fully data-driven latent variable models — then uses that foundation in the afternoon to work through the strategic and organizational consequences. The goal is to equip participants to make informed decisions about ML integration in their own pipelines, not to advocate for a single approach.

Instructor: Niklas Korsbo, PhD

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

Agenda
Opening: What This Is Actually About (20 min)
The pressures driving interest in ML for pharmacometrics — workflow efficiency, decision quality, cross-program learning — and what decision-makers need to understand versus what they can delegate.
The Modeling Spectrum
The random effect, reformulated. How neural networks embed into dynamical systems. Parallels between random effects and latent variables in generative models. A unified view through the marginal likelihood — showing that mechanistic NLME and deep learning target the same objective under different constraints.

Hybrid Strategies That Work Today

• Mixing mechanistic knowledge with data-driven neural networks in DeepNLME.
• Ensemble approaches: mechanistic models for primary endpoints, flexible models for biomarkers.
• Post-hoc joining of independently trained models, enabling information transfer and modular model platform building.
• Why fully data-driven modeling isn't optimal today — and how that might evolve.

From Models to Platforms
How increased flexibility enables translation across datasets, doses, and indications. Fine-tuning as a bridging mechanism between programs. Multi-biomarker integration, multi-modal data, and the progression toward latent-space reasoning across an indication area.

The Technology Stack and the Decisions It Forces
How ML capabilities in pharmacometrics build on each other — and where value is generated at each stage. From single-model hybrids through multi-endpoint integration to cross-program learning and multi-modal data. What each level delivers concretely, what it requires in infrastructure and data, and how partial deliverables justify the next investment. Crucially, some of the choices that determine whether you can use these methods in three years — data collection practices, skill development, infrastructure — need to be made now, before the capability is fully mature. What to invest in anticipation versus what to wait on, and how to distinguish bets worth making from premature commitments.
The Bigger Picture: What Is Pharmacometrics For?
If these capabilities mature, the role of the pharmacometrics department changes fundamentally. Today, PMx primarily provides linear trial decision support — de-risking and optimizing one program at a time. But a department that can learn across programs, integrate multi-modal data, and build reusable modeling platforms becomes something different: a validator of preclinical work, a closer of the feedback loop into the next cycle of drug development, and a critical informer of post-market strategy. From service function to platform builder. What does that look like in practice, and what would it take?
Where This Is Heading
A five-year view — with honest acknowledgment of what remains uncertain. What you can act on now. Open questions for the field.
Instructional Approach
The morning is instructor-led, building shared technical language. The afternoon is exploratory: frameworks and scenarios are presented, and participants work through implications for their own organizations. The workshop is technically grounded but aims at strategic intuition — mathematical fluency is not assumed.
About the Instructor
Niklas Korsbo leads development of DeepPumas at PumasAI, building the tooling that integrates deep learning into nonlinear mixed-effects modeling for drug development. He advises pharmaceutical partners on ML strategy for pharmacometrics and has delivered workshops at PAGE, ACoP, and on-site at major pharma organizations. His work sits at the intersection of applied pharmacometrics and modern machine learning — he builds the methods, applies them in real programs, and teaches others to evaluate them.

Pricing:
Early bird pricing is available until September 1.

• ISoP Member: $795 early bird / $895 regular
• Non-Member: $895 early bird / $995 regular
• Student/Trainee: $250 early bird / $300 regular

Bring Your Team, Save More! Groups of 3+ qualify for a group discount. Email [email protected] to inquire.