Date: June 2-3, 2025

Location: PAGE 2025

Thessaloniki Concert Hall, 54646 Thessaloniki Greece

Register now!

DeepPumas is a powerful tool that bridges the gap between mechanistic, statistical, and machine learning modeling. This hands-on workshop will introduce participants to the core concepts behind DeepPumas and its potential to transform decision-making in the healthcare space. By seamlessly embedding machine learning methodology into nonlinear mixed-effect models, DeepPumas enables the simultaneous utilization of domain-specific knowledge and available data for model identification.

In this workshop, we will learn about NeuralODEs, universal differential equations and scientific machine learning, and how they can be leveraged to discover unknown dynamics driving disease progression and treatment response. We will then see how DeepNLME expands these concepts to include random effects. This lets us discover individualizable models that can tell us something not only about the measurements in our data but also about the patients whose outcomes we are trying to understand. Furthermore, we will learn how to leverage machine learning to identify useful prognostic factors from complex data such as images or free-form text. After gaining both theoretical understanding and practical experience of using these new technologies, we will explore different use cases to get a sense of how DeepPumas is poised to transform the role of data and modelling in healthcare.

Instructors: Niklas Korsbo, PhD and Patrick Mogensen, PhD

Learning Objectives:

1. Learn basic usage of the Pumas and DeepPumas software.
2. Understand simple neural networks and how to use them.
3. Understand neural-embedded dynamical systems such as Neural Ordinary Differential Equations and Universal Differential Equations.
4. Understand random effects, their connection to latent spaces of generative AI, and how they can be leveraged to create powerfully individualizable models for longitudinal data.
5. Learn how to integrate machine learning for prognostic factor identification in longitudinal models.
6. Understand how to compose all the above to enable a semi-mechanistic scientific machine learning approach to finding individualizable and predictive longitudinal models.
7. Get a sense of how this technology may transform the role of data and modeling in the healthcare space.

Register now!