In silico medicine is redefining how we understand, predict, and optimize health interventions. By leveraging mechanistic models, it is now possible to simulate disease dynamics and therapeutic responses entirely in silico—offering a powerful complement to traditional experimentation. When combined with AI-based analytics for knowledge extraction and data integration, these models can further enhance our capacity to represent complex biological systems and personalize medical interventions.

This keynote will explore the evolving landscape of in silico medicine and its growing recognition by regulatory bodies. Drawing on recent developments within the European Medicines Agency (EMA) and international initiatives, I will discuss what it truly means for a computational model to reach regulatory maturity, emphasizing validation, transparency, and contextual credibility as key pillars of digital evidence acceptance.

Through examples from EU-funded programs and real-world regulatory interactions, the talk will illustrate how AI can effectively support mechanistic modeling—accelerating model development, parameterization, and hypothesis generation—while maintaining the scientific interpretability essential for regulatory use.

Ultimately, the aim is to chart how in silico medicine is progressing from scientific innovation to a trusted regulatory asset, driving the next generation of predictive and personalized healthcare.

The convergence of spatial transcriptomics and high-resolution digital pathology imaging has revolutionized our understanding of tissue architecture and disease biology. This talk explores the integration of these powerful modalities, enabling a spatially resolved analysis that bridges engineering with basic and translational science. By harnessing generative AI, we can now simulate virtual cells and tissue systems, offering unprecedented opportunities for in silico experimentation and discovery. This synergistic approach not only deepens our insights into the complex interplay of molecular and morphological data but also paves the way for transformative advancements in precision medicine and therapeutic development.

In this talk, I will present a brief overview of the work that myself and others have done to discover disease-relevant biomarkers in the T-cell receptor repertoires from patients. I will detail why immune repertoires have such a large potential to change the way we diagnose diseases, and what hurdles still need to be overcome for them to be applied in daily clinical use. A primary focus of the talk will be the recent AI solutions that have been developed to overcome the intrinsic uniqueness of these repertoires, and how they have lead to novel disease insights and solutions for therapies and diagnostics.

Just as large language models (LLMs) transform human language into computational understanding, nature has already encoded a deeper logic—an immune language composed of receptor sequences, co-occurrence patterns, and network structures shaped by evolution. This language does not predict words. It predicts survival.

We present AIDeN—the Adaptive Immune Defensive Network—a biologically trained model that functions as a consensus reference for the adaptive immune system, much like the reference genome transformed genomics. Built from population-scale immune repertoire data and designed with network-based AI logic, AIDeN captures the structural memory of immunity across individuals.

Instead of isolated biomarkers, AIDeN learns how receptor “nodetypes” co-occur in patterns—called “linklets”—that define immune grammar. Each repertoire becomes a network fingerprint; deviations from the consensus reveal early dysfunction—before disease is clinically evident.

For pharmaceutical development, AIDeN offers new tools to enhance trial design and accelerate drug discovery. It enables:

• Identification of responders based on immune network fitness,
• Discovery of new immunological targets through structural deviation mapping,
• Faster, quantitative endpoints by tracking immune rewiring over time.

AIDeN shifts the paradigm from symptom-based classification to network-informed precision. It supports diagnostics, monitoring, and therapy optimization by evaluating immune system integrity at scale.

Where LLMs decode culture, AIDeN decodes survival. It is the immune system’s own model—built by nature, revealed by data, and ready to guide the future of precision immunology and immune-guided therapeutics.

In this presentation, I will provide an overview of our antibody discovery platform, highlighting how it integrates multiple computational technologies to streamline and optimize the discovery process. We have developed a suite of digital tools spanning traditional bioinformatics techniques, high-throughput data analysis workflows, and advanced ML/AI-driven models to address the complex challenges inherent in antibody screening and selection. By employing these methods, we analyze and refine candidate pools, prioritizing promising antibody clones based on structural, functional, and biological parameters. Our digitally assisted screening improves and accelerates the identification of top-performing clones and offers deeper insights into their underlying mechanisms of action. While we have made significant progress in improving selection accuracy and reducing development timelines, several challenges remain, particularly in data quality, model interpretability, and the ongoing evolution of AI methodologies. To tackle these challenges, we foster close collaborations among scientists, data scientists, and computational biologists, enabling us to continuously refine our toolbox and adapt to emerging needs. We are currently exploring novel paradigms such as generative models and active learning to further expand antibody space exploration. Through this presentation, I aim to share insights from our experiences building and deploying digital solutions for antibody discovery, highlight some of the key breakthroughs achieved, and outline the opportunities that still lie ahead.

Digital Twins (DTs) have transformative potential for drug discovery and, in particular, for clinical trials. DTs can support clinical trial decision-making and address challenges such as restrictive inclusion/exclusion criteria and virtual control arm simulation. Here we present an LLM-based DT for state-of-the-art forecasting of patient clinical trajectories that we have developed by leveraging electronic health records (EHRs). Advantages of using LLMs for forecasting include native handling of missingness and noise, as well as explainability through a human-interpretable interface. Further development includes robust evaluation strategies, investigation of model interpretability, and progression towards regulatory approval, with a focus on clinical impact.

Andreas Hess, Jürgen Rech, Arnd Dörfler, Georg Schett, Mageshwar Selvakumar, Germany

Diego Diaz-Milanes, Francesa Bureca, Raquel Remesal-Cobreros, Spain

Mehdi Shojaei, Bjoern Eiben, Jamie McClelland, Simeon Nill, Alex Dunlop, Arabella Hunt, Brian Ng-Cheng-Hin, Uwe Oelfke, United Kingdom

Jing Zhai, Sergey Oreshkov, Nelly Pitteloud, Federico Santoni, Switzerland