COLIBRI Doc Day

Program

09:00-09:20 Coffee & welcoming words

Session 1 (Chair: Federica Caforio, Uni Graz)

9:30: Christoph Zechner, SISSA Trieste  
Bayesian inference of chromatin and transcription dynamics in living cells
10:20: Duc-Anh Nguyen, Uni Graz 
Global well-posedness and stability analysis of a degenerate reaction-diffusion system for a prey-predator and natural enemy interaction
10:40: Birgit Tschiatschek, Uni Graz
Guardians of neutrality: stem cell competition and tissue homeostasis

11:00-11:20: Coffee break

Session 2 (Chair: Adrián Aguirre-Tamaral, Uni Graz)

11:20: Jan Korbel, Complexity Science Hub Vienna 
Stochastic Thermodynamics of Information and Computation
12:10: Fabian Veider, Uni Graz
Clustering in Adaptive Voter Models: An enhanced equation-based approach

12:30-14:00 Lunch (catering on-site)

Session 3 (Chair: Lara Trussardi, Uni Graz)

14:00: Federica Caforio, Uni Graz
Integration of physics-informed machine learning and mathematical modelling for cardiac digital twins
14:50: Laurenz Fedotoff, Uni Graz
Using Expert Knowledge for Feature Engineering in Multimodal Neural Network Development
15:10: Angelika Manhart Uni Wien
Why use Mathematical Modelling in Cell Biology? Tales of Nuclear Positioning & Size Scaling

16:00-16:15: coffee break

16:15-17:30: Doctoral forum
 

Abstracts

Bayesian inference of chromatin- and transcription dynamics in living cells
Christoph Zechner (SISSA, Trieste)
Recent live-cell microscopy techniques allow the simultaneous tracking of distal genomic elements and transcriptional activity, offering new ways to study chromatin dynamics and gene regulation. However, drawing robust conclusions from such data is statistically challenging due to substantial technical noise, intrinsic fluctuations and limited time resolution. In this talk I will discuss recent progress we have made in addressing these challenges. First, I will present a statistical method to quantify CTCF/cohesin-mediated chromatin looping dynamics from two-point live-cell imaging experiments. The method combines a simple polymer model with Bayesian filtering to infer loop lifetimes and frequencies. Its application to experimental measurements of the Fbn2 loop in mouse embryonic stem cells revealed that chromatin loops are surprisingly rare (~5% looped fraction) and short-lived (~20mins loop lifetime). I will then discuss how simultaneous transcriptional readouts (MS2/PP7) can be integrated with chromatin tracking data to quantify enhancer-promoter communication and its effect on transcription. I will conclude by highlighting potential implications of such approaches for understanding gene regulation in single cells and outline current and future challenges.

Global well-posedness and stability analysis of a degenerate reaction-diffusion system for a prey-predator and natural enemy interaction
Duc-Anh Nguyen (Uni Graz)
In this study, we introduce a degenerate reaction-diffusion system that models interactions among prey, predator, and the predator's natural enemy. A typical scenario arises in crop-pest-natural enemy interactions in agriculture, in which the crop is a plant species that lacks diffusion; the natural enemy not only feeds on the predator but also supports the crop's growth. Firstly, the well-posedness properties, such as global existence and uniform boundedness in time of the solution, are proved. Next, we present the stability analysis of the constant equilibria, including linear and nonlinear stability.

Guardians of neutrality: stem cell competition and tissue homeostasis
Birgit Tschiatschek (Uni Graz)
Adult tissue renewal relies on stem cells to replace old or damaged cells, thereby maintaining tissue homeostasis. While the hierarchical model of stem cell organization has long been the prevailing paradigm, recent studies suggest that stem cell behavior in several tissues is governed by neutral competition. In this stochastic process, equipotent cells randomly function as stem cells, challenging the traditional view of deterministic stem cell behavior.
If cell fate is stochastic, a fundamental question emerges: what keeps stem cell competition neutral? Focusing on the intestine and pancreas, we investigate mechanisms that preserve balanced stem cell dynamics and ensure tissue stability. These "guardians of neutrality" operate through diverse mechanisms that may vary across tissues, but collectively support homeostasis. Understanding these mechanisms provides insight into both tissue maintenance and disease development.

Stochastic Thermodynamics of Information and Computation
Jan Korbel (Complexity Science Hub Vienna)
Computation is usually treated as an abstract manipulation of symbols, but every real computer is a physical system in contact with its environment. This talk will introduce how stochastic thermodynamics connects information processing, irreversibility, and energetic cost. Starting from the second law and Landauer’s principle, we will discuss why erasing information has a thermodynamic price in terms of dissipated energy, why real systems operate far above this bound, and how correlations, initial distribution, speed, precision, and architecture affect the cost of computation in artificial and natural systems. 

Clustering in Adaptive Voter Models: An enhanced equation-based approach
Fabian Veider (Uni Graz)
The rise of social media and algorithmic recommendation systems raises concerns about their role in amplifying opinion polarization and fostering echo chambers. To understand how network structure shapes these dynamics, we analyze an adaptive voter model based on common social influence theory such as opinion imitation and link adjustments. The model shows a critical transition between consensus and fragmentation based on the link rewiring rate, which is captured qualitatively by moment-based approaches but lacks quantitative agreement at the transition point. In order to mitigate this mismatch, we go beyond the standard pair approximation by applying a triangle expansion scheme that incorporates the total clustering in our equation-based approach. This accounts for closed-triangle contributions and yields an improved analytical expression for the critical rewiring rate. Both the pair approximation and our extension overestimate the critical transition point at which the network fragments into disconnected components, but our model substantially reduces the discrepancy and brings predicted behavior into much closer agreement with simulations while still being analytically tractable. We validate our estimate of critical transitions on networks with diverse structural properties, including Erdős-Rényi, Watts-Strogatz, and Barabási-Albert networks, finding improvements with simulations particularly in moderately to highly clustered regimes. We further find improved agreement of steady-states and compare discrepancies between time evolutions of simulations and our analytical treatment. Our results show that incorporating clustering into moment-based descriptions can significantly improve predictions of fragmentation transitions, advancing the theoretical understanding of critical transitions in adaptive voter models and their implications for opinion polarization in different social networks.

Integration of physics-informed machine learning and mathematical modelling for cardiac digital twins
Federica Caforio (Uni Graz)
Cardiac biophysical models are advancing rapidly owing to their predictive capabilities, yet the computational cost of high-resolution multi-physics models and their personalisation remains a barrier to clinical translation. This work introduces a methodology integrating physics-informed neural networks (PINNs) with 3D, time-dependent nonlinear biomechanical models of cardiac tissue to reconstruct displacement fields and estimate heterogeneous, patient-specific biophysical properties. The learning algorithm uses sparse displacement and, where available, strain data that can be routinely obtainable in clinical settings, and incorporates residual-based attention, Fourier features, and tailored regularisation strategies to enable robust parameter field reconstruction under noisy conditions at high spatial resolution. A Pareto front analysis is performed to assess the influence of loss weight selection on parameter estimation. Several benchmarks demonstrate the accuracy and robustness of the method, showing that PINNs can detect the presence, location, and severity of scar tissue, thus offering potential improvements to diagnostics and treatment planning for cardiac disease. Recent advances in WarpPINN-based approaches for extracting displacement fields from cine MRI are also discussed.

Using Expert Knowledge for Feature Engineering in Multimodal Neural Network Development
Laurenz Fedotoff (Uni Graz)
The honey bee queen is the central individual in the complex superorganism of the honeybee colony, solely responsible for all offspring and colony growth, making her the source of the immense ecosystem service provided by the workers. The field of ethology is shifting from hypothesis-driven to data-driven research, demanding AI capabilities for analyzing large datasets produced. In the project ROBOROYALE 6 TB of high-resolution image data are collected daily, focusing on the queen's behavior. To analyze this data, I am developing specialized multimodal AI systems, using 2D Convolutional Neural Networks, to extract novel insights about the queen's long-term behavioral profile.

Why use Mathematical Modelling in Cell Biology? Tales of Nuclear Positioning & Size Scaling
Angelika Manhart (Uni Wien)
For the COLIBRI Doc Day, I would like to share my experiences working at the interface between mathematics and biology. To ground this in science, I will focus on a topic that has accompanied me throughout my academic career: the cell nucleus. I will show how, together with collaborators, we used model-based hypothesis testing to understand the mechanisms that determine nuclear size and position in both single- and multinucleated cells. The idea is to not formulate one computational model, but a whole family of models, and then use data to identify the most plausible model - and therefore the most plausible biological mechanism. Through these examples, I would like to demonstrate one way how mathematical modelling can create biological insight.