Informatik Festival 2026

The transformative dynamics of generative artificial intelligence, particularly large language models, are increasingly influencing core processes of academic practice. Alongside substantial efficiency gains, these technologies are prompting a re-evaluation of established practices in knowledge generation, dissemination, and infrastructure. This paper examines the disruptive effects of this development using a simplified model of academic work comprising three dimensions: research, teaching, and information supply. At its core, the talk critically analyzes whether the observed challenges represent genuinely novel phenomena or primarily act as catalysts for pre-existing systemic vulnerabilities.

In research, these mechanisms can be illustrated through automated literature synthesis and data-driven pipeline creation. While these methods drastically accelerate research and analysis processes, they simultaneously amplify well-known misaligned incentives. The simplified generation of scientific texts feeds into the existing pressure to publish (‘publish or perish’) without necessarily yielding a qualitative increase in scholarly insight.

In the realm of teaching, the trivialized automated generation of academic texts and program code reveals a structural shift: the decoupling of the creation process from the final product. Taking traditional forms of assessment – such as term papers or programming assignments – as examples, it becomes clear that the finished artifact can no longer unequivocally serve as an indicator of cognitive learning gain. This challenges higher education didactics to fundamentally redefine assessment formats and learning objectives.

A key focus of the talk lies in the field of information supply, where the transformation is illustrated through the case of unlocking historical sources. While conventional OCR and HTR systems reached methodological limits with complex manuscripts, multimodal language models achieve substantial leaps in quality regarding transcription and contextualization. However, this gain in accessibility is counterbalanced by novel quality assurance challenges: synthesized, visually plausible, yet hallucinated outputs and references impede traceability, error identification, and provenance verification.

For academic libraries and infrastructure organizations, the technological shift implies the potential of a general functional realignment – moving from the primary provision of documents toward fostering critical validation skills and curating reliable data spaces.

The talk is primarily intended to serve as a catalyst for discussion. It problematizes the need to adapt academic standards, evaluation mechanisms, and information infrastructures to changing technological conditions without oversimplifying cause and effect.

Computational biology is an interdisciplinary field with contributors from a range of backgrounds. Reproducibility therefore hinges not only on technical factors but also on the usability of the tools involved. In this paper, we present a test of the MorpheusML ecosystem in which three teams of high school students were invited to develop MorpheusML models for biological processes as described in provided papers and to submit their models to the public MorpheusML model repository via a structured publishing process. All three teams achieved and submitted a working model within a single workday. In the first case study, the students additionally uncovered a parameter swap in an influential paper, underlining the value of reproduction as a quality-control mechanism. We conclude that the MorpheusML ecosystem is usable in practice and well-suited as both a reproducibility and an educational tool.

The development of virtual human twins requires computational models that are mechanistically detailed, yet also reproducible and extensible. In liver biology, diverse multicellular models have been proposed to describe processes such as zonation, regeneration, and toxic injury, but differences in implementation limit their comparability and reuse. Here, we systematically reimplement representative liver models within the unified simulation framework Morpheus using its declarative language MorpheusML.This enables consistent integration of spatial cell-based dynamics, intracellular signaling, and extracellular transport. We show that key biological behaviors can be reproduced across models while revealing sensitivities to parameterization and structural assumptions. Importantly, we demonstrate that MorpheusML supports model reuse and composability, allowing existing models to be modified and combined to generate new insights. Our results highlight the value of standardized modeling frameworks as a foundation for integrated, multiscale simulations, representing a key step toward predictive virtual human twins.

Muscle regeneration after injury requires a balance between the proliferation of activated muscle stem cells and their differentiation into myocytes. These cell-fate decisions are controlled by intracellular gene expression, integrating extracellular signals. Core elements of the intracellular regulatory network are the transcription factors Hes1 and MyoD, whose expression dynamics have been shown to be linked to different cell states. While oscillatory protein expression is linked to proliferation, sustained protein expression is linked to differentiation. The integrated extracellular signal depends on the direct environment of the cell. Cells that share a boundary communicate via the Delta–Notch pathway and thereby couple their intracellular regulatory networks through activation of Hes1 expression. Additionally, Hes1 is activated by soluble growth factors recognized over the outer membrane.

Although the individual processes have already been studied, it remained unclear whether this known regulations and interactions can coordinate the cell fate decisions in time and space that a single activated muscle stem cell can repopulated an initially cell-free region with differentiated myocytes?

Using Morpheus, we developed a spatial 2D model of myocyte generation after muscle stem cell activation. Individual cells are represented by a Cellular Potts Model and contain a system of delayed differential equations (DDE) describing the dynamics of Hes1, Dll1 and MyoD and the transition between proliferating and differentiated cell states. The DDEs include inputs from the surrounding environment, i.e. Dll1 from neighboring cells or growth factor from the cell-free space, weighted by the respective shared boundary length. Stochastic cell movement and state-dependent cell division dynamically alter these inputs, coupling spatial cell dynamics to intracellular regulation

Simulations show that spatial cell arrangement and cell movement can strongly affect Hes1 dynamics as well as proliferation and differentiation rates. The arising spatial pattern of differentiation and its dynamics depend on the extent and the timing of cell–cell contact. Systematic variation of the strengths of growth-factor and Notch-mediated Hes1 activation revealed parameter regions characterized by premature termination of proliferation, balanced proliferation and differentiation, or overgrowth. Furthermore reduced activity in one Hes1 activation mode can be partially compensated by the increased activity of the other mode.

The model demonstrates how changes in cell arrangement can alter intracellular dynamics and cell-fate decisions without requiring changes in the intracellular regulatory network itself. It provides a framework for studying how cell movement, proliferation and contact-dependent signaling jointly affect muscle regeneration.

Haematopoietic stem cell (HSC) maintenance in the bone marrow (BM) – encompassing self-renewal, differentiation, proliferation, and quiescence – depends critically on the local microenvironment, the so called stem cell niche. Growing evidence indicates that niche dysregulation contributes to haematologic malignancies. However, the intricate spatial architecture of the BM hampers a comprehensive understanding of niche-driven HSC regulation and its role in disease progression.

To overcome this, we are developing a 3D spatio-temporal in silico model within the simulation environment Morpheus. In this framework, HSCs and progenitor cells are represented as individual agents whose movement and behaviour evolve over time through deterministic and stochastic rules. These agents interact with each other and with niche components (e.g. endothelial cells, mesenchymal stromal cells, or diffusible signalling molecules) within a 3D domain reflecting niche geometry – derived either directly from microscopy data or generated via a Gierer-Meinhardt reaction-diffusion system.

Currently, the model incorporates two reversible HSC states: ‘active‘ (proliferating) and ‘inactive’ (quiescent), extending a previous non-spatial model into a spatial context. Preliminary simulations in artificially generated scaffolds demonstrate that HSC population dynamics and residence time distributions are strongly influenced by the scaffold structure. These results suggest that structural alterations in the niche can modulate HSC behaviour, offering a potential mechanism for age- or disease-related microenvironmental changes.

We adopt an iterative modelling approach, continuously validating simulation outputs against experimental data and refining the model accordingly. This strategy aims to advance mechanistic understanding of spatial niche regulation in haematopoiesis and its dysregulation in aging and disease.

NF-κB signaling shapes the inflammatory response, and its dysregulation is linked with autoimmune, neurodegenerative, and cardiovascular diseases. After pathway activation by a large variety of stimuli, dimers of the NF-κB family act as transcription factors for a diverse set of target genes including positive and negative regulators of the upstream pathway. A special case is the cytokine TNFα, which is not only transcribed in a NF-κB dependent manner but also secreted and results in NF-κB activation via the TNFR receptor cascade in neighboring cells. For example, macrophages in the intestinal crypt respond to bacterial metabolites, such as LPS, with TNFα secretion and alert surrounding immune cells to invasion using this mechanism.

We utilize a computational model of canonical NF-κB signaling activated by TNFα and LPS to examine the impact of spatial properties on cell-cell communication. Starting from an established model for the intracellular dynamics of canonical NF-κB signaling, we expand into a model coupling cells via TNFα. To investigate the interactions between different NF-κB expressing cell types, a subset of cells is stimulated by LPS to secrete TNFα which induces a signaling cascade in the surrounding cells, mirroring interactions found in the intestinal crypt.

Using computational analysis, we identify which model parameters, and their corresponding biochemical processes, have the strongest impact signal transmission speed and loss of signal.

Reproducibility is a cornerstone of science, yet many models in computational biology cannot be re-executed years after publication – whether due to lost source code, unmaintained dependencies, or undocumented modelling choices. We present the resurrection of an influential multiscale model of vertebrate somitogenesis by Hester et al., whose source code was no longer available. From the publication alone, we reconstruct the full model – a Cellular Potts simulation coupled to a 20-ODE intracellular clock and an FGF wavefront – within the Morpheus ecosystem. Qualitative and quantitative comparisons against the published results show that our reconstruction reproduces the original behaviour indistinguishably. Distributed as a FAIR-compliant MorpheusML file via the MorpheusML Model Repository, the model is now readily accessible for future research.