Giersch International Conference & SCALE Kick-Off

2 Mar 2026, 09:00 → 5 Mar 2026, 19:15 Europe/Zurich

Lecture Hall (FIAS / OSZ)

Sebastian Thallmair (Frankfurt Institute for Advanced Studies), Roberto Covino, Michaela Müller-McNicoll

The Cluster of Excellence SCALE – SubCellular Architecture of LifE  aims to build digital models of human cellular segments in molecular detail, allowing them to evolve in virtual reality through advanced simulations. To launch the Cluster, we organize a Kick-Off jointly with the Giersch International Conference entitled "Towards Digital Twins for Structural Cell Biology - Criteria, Chances and Challenges”.

The Giersch International Conference series at FIAS (Frankfurt Institute for Advanced Studies) was established in 2016 and generously funded by the Giersch Foundation. The conference series focuses on interdisciplinary topics, bringing together leading experts to discuss with young scientists at FIAS.

Digital Twins for Structural Cell Biology – Criteria, Chances and Challenges

Originally emerging from industry and engineering, digital twins are now set to enter the field of natural sciences. With their aim to offer real interrogability akin to true experimental conditions, but with much higher controllability and reproducibility, digital twins promise to overcome existing experimental frontiers and to facilitate mechanistic insights to both theorists and experimentalists. The ongoing integration of classical simulation techniques and more recent AI-based approaches further enhances these innovative opportunities.

However, creating digital twins of natural systems is challenging due to their inherent complexity and stochasticity. Unlike in industrial applications, where components are well characterized, the structure and function of natural system components are often hardly known and itself subject to investigation. Moreover, creation of numerically efficient digital twins requires careful, system-specific abstraction of natural complexity while retaining mechanistic interrogability. These difficulties call for new approaches and a thorough balancing of existing methodology.

This conference will bring together leading experts engaging in digital twin development for structural cell biology. We will identify challenges and concepts in digital twin implementation and attempt to establish a comprehensive catalog of criteria for the next generation of high-quality digital twins for structural cell biology – an important goal of the Cluster of Excellence SCALE.

Topical sessions:

- Computational Approaches for Digital Twins
- Digital Twins of Subcellular Segments
- Digital Twins of Multicellular Systems
- Digital Twins of Neural Systems

Program Giersch Conference & SCALE kick off 2026.pdf

Short Schedule Giersch SCALE Conference.pdf

Monday 2 March

Welcome: Registration and Coffee

Welcome: Opening Remarks

Invited Talks

1 Toward Digital Twins of Development and Disease

Computational simulations have long been used to study emergent phenomena in biology. Data-driven in silico models of tissue behavior in development and disease now enable the creation of Digital Twins—virtual counterparts with applications in bioengineering and precision medicine. In this talk, I will present our simulation frameworks for high-resolution tissue modeling and parameter inference from imaging and experimental data. Finally, I will discuss our progress in constructing Digital Twins for epithelial tissues, morphogenesis, and medical treatments.

Speaker: Prof. Dagmar Iber (ETH Zürich)

2 Integrative Structural Biology in the era of Artificial Intelligence

Speaker: Dr Max Bonomi (Institut Pasteur)

Welcome: Reception

Tuesday 3 March

Invited Talks

3 How cellular architecture modulates drug response

Speaker: Prof. Stefan Knapp (Goethe University)

4 Bayesian metamodeling of early T-cell antigen receptor signaling accounts for its nanoscale activation patterns

T cells respond swiftly, specifically, sensitively, and robustly to cognate antigens presented on the surface of antigen presenting cells. Existing microscopic models capture various aspects of early T-cell antigen receptor (TCR) signaling at the molecular level. However, none of these models account for the totality of the data, impeding our understanding of early T-cell activation. Here, we study early TCR signaling using Bayesian metamodeling, an approach for systematically integrating multiple partial models into a metamodel of a complex system. We inform the partial models using multiple published super-resolution microscopy datasets. Collectively, these datasets describe the spatiotemporal organization, activity, interactions, and dynamics of TCR, CD45 and Lck signaling molecules in the early-forming immune synapse, and the concurrent membrane alterations. The resulting metamodel accounts for a distinct nanoscale dynamic pattern that could not be accounted for by any of the partial models on their own: a ring of phosphorylated TCR molecules, enriched at the periphery of early T cell contacts and confined by a proximal ring of CD45 molecules. The metamodel suggests this pattern results from limited activity range for the Lck molecules, acting as signaling messengers between kinetically-segregated TCR and CD45 molecules. We assessed the potential effect of Lck activity range on TCR phosphorylation and robust T cell activation for various pMHC:TCR association strengths, in the specific setting of an initial contact. We also inspected the impact of localized Lck inhibition via Csk recruitment to pTCRs, and that of splicing isoforms of CD45 on kinetic segregation. Due to the inherent scalability and adaptability of integrating independent partial models via Bayesian metamodeling, this approach can elucidate additional aspects of cell signaling and decision making.

Speaker: Dr Barak Raveh (The Hebrew University of Jerusalem)

10:30 Coffee Break

Short Talk

5 From in situ cryoET to digital twins of subcellular segments

High-confidence 3D template matching (hcTM) turns in situ cryoET tomograms into simulation-ready subcellular segments. Multiscale simulations then make predictions later confirmed experimentally. For example, how mutations or composition changes propagate to mesoscale organization and biological function. Examples include human chromatin organization, viral transport and mutation-driven condensate solidification.

Speaker: Dr Sergio Cruz-Leon (Max Planck Institute of Biophysics)

6 Control of the condensation of TDP-43 by enzymatic phosphorylation: a perspective from molecular dynamics simulations

Cellular processes are organized by the phase separation of proteins into biomolecular condensates. These condensates are regulated by post-translational modifications, most notably phosphorylation. Phosphorylation of proteins is catalysed by kinases which consumes the chemical fuel ATP. While the phosphorylation of TDP-43 is closely linked to neurodegenerative disease, how TDP-43 is phosphorylated remains poorly understood. It is not clear whether kinases interact directly with condensates or primarily with TDP-43 proteins within the dilute phase. Molecular dynamics simulations can potentially resolve these interactions, yet modelling fuel-driven, non-equilibrium dynamics remains a significant computational challenge. We show how to simulate chemical-driven enzymatic phosphorylation of proteins in coarse-grained molecular dynamics. Importantly, we demonstrate how to validate the thermodynamic consistency of these simulations by automatically constructing Markov-state models from the resulting trajectories. Our results reveal that the kinase Casein kinase 1 delta binds preferentially to TDP-43 condensates over the dilute phase, thereby accelerating phosphorylation. While enzymes initially localise to the droplet surface, phosphorylation enables enzymes to interact with the condensate interior. Our simulations demonstrate that ultimately this localised enzymatic activity can trigger the complete dissolution of TDP-43 condensates.

Speaker: Prof. Lukas Stelzl (Johannes Gutenberg University Mainz & IMB)

Invited Talks

7 Computational tools for building models of entire biological systems from light and electron microscopy

Speaker: Stephan Preibisch (HHMI Janelia)

12:30 LUNCH BREAK

Invited Talks

8 From integrative structural biology to cell biology

Integrative modeling is an increasingly important tool in structural biology, providing structures by combining data from varied experimental methods and prior information. As a result, molecular architectures of large, heterogeneous, and dynamic systems, such as the ~52 MDa Nuclear Pore Complex, can be mapped with useful accuracy, precision, and completeness. Key challenges in improving integrative modeling include expanding model representations, increasing the variety of input data and prior information, quantifying a match between input information and a model in a Bayesian fashion, inventing more efficient structural sampling, as well as developing better model assessment, analysis, and visualization. In addition, two community-level challenges in integrative modeling are being addressed under the auspices of the Worldwide Protein Data Bank (wwPDB). First, the impact of integrative structures is maximized by PDB-Dev, a prototype wwPDB repository for archiving, validating, visualizing, and disseminating integrative structures. Second, the scope of structural biology is expanded by linking the wwPDB resource for integrative structures with archives of data that have not been generally used for structure determination but are increasingly important for computing integrative structures, such as data from various types of mass spectrometry, spectroscopy, optical microscopy, proteomics, and genetics. To address the largest of modeling problems, a type of integrative modeling called metamodeling is being developed; metamodeling combines different types of input models as opposed to different types of data to compute an output model. Collectively, these developments will facilitate the structural biology mindset in cell biology and underpin spatiotemporal mapping of the entire cell.

Speaker: Dr Andrej Sali (UCSF)

9 Biomolecular condensate architecture of an autophagic cargo at molecular resolution in situ

Biomolecular condensates organise cellular biochemistry, yet their molecular architecture in situ remains poorly understood. During selective autophagy, macromolecules frequently accumulate into biomolecular condensates, forming discrete entities for autophagic engulfment and degradation – ideal systems for structural analysis. We employed in situ cryo-electron tomography to determine the near-atomic resolution structure of Aminopeptidase 1 condensates within cells. These condensates form densely packed, spherical assemblies with amorphous organisation and liquid-like properties, elucidating the requirements of a selective autophagic cargo for exclusive targeting. Structural analysis and multiscale simulations reveal that the short, transient α-helical structures in the disordered N-terminus of Aminopeptidase 1 enable site-specific, coiled-coil-like interactions required for condensate formation and properties. A single point mutation that increases α-helical propensity directly modulates condensate viscosity and dynamics from a liquid-like to a glass-like state, while preserving local molecular packing. Our results demonstrate that disordered regions encode both specificity and material properties through transiently structured motifs, linking sequence specificity to phase behaviour in cells and expanding the molecular logic of phase separation in cells.

Speaker: Dr Florian Wilfing (Max Planck Institute of Biophysics)

15:30 Coffee Break

Short Talk

10 Multiscale Modeling of Nuclear Membrane Sealing

During mitotic exit, the nuclear envelope seals spindle‑microtubule holes in a process mediated via LEM2‑ESCRT. Cryo-electron tomography (Cryo-ET) shows LEM2-ESCRT filament architecture but not the sealing mechanism. We built a digital twin by integrating Cryo‑ET with multiscale molecular dynamics simulations of filaments in a 90 nm membrane hole, capturing conformational changes, showing how LEM2’s disordered domain engages microtubules, and allowing in‑silico tests of mechanical stress.

Speaker: Dr Karen Palacio-Rodriguez (Max Planck Institute of Biophysics)

11 Deep-SegCLEM: unified segmentation and matching for correlative light–electron microscopy

Correlative light and electron microscopy (CLEM) links dynamic functional imaging with ultrastructural detail, yet automated correlation remains unresolved due to the difficulty of bridging two fundamentally different microscopy modalities. Current approaches either bypass segmentation or rely on generic models, leading to poor or inconsistent alignment. We present Deep-SegCLEM, a fully automated pipeline that leverages mitochondria-specific features to establish fiducial-free correspondence between LM and EM images. Deep-SegCLEM integrates two tailored segmentation networks—DSCLEM-LM for fluorescence microscopy and DSCLEM-EM for electron microscopy—with a multi-scale template-matching algorithm for cross-modal alignment. On fluorescence datasets spanning confocal, Airyscan, and SIM modalities, DSCLEM-LM achieved state-of-the-art performance (F1 = 0.98, IoU = 0.96), surpassing U-Net, U-Net++, DeepLabV3, FCNN, MitoSegNet (F1 = 0.93), and the μSAM foundation model (F1 = 0.73). DSCLEM-EM similarly outperformed classical CNN architectures and foundation models across multiple public EM benchmarks, including challenging conditions such as UroCell (F1 = 0.80 versus ≤0.57 for CNN baselines and ≤0.72 for μSAM). Failure-Driven Targeted Augmentation (FDTA) was essential for capturing rare morphologies and improving generalization. The correlation module achieves fully unsupervised LM–EM alignment in minutes, with over 80% matching accuracy and a mean centroid error of 2.0 μm relative to expert-generated STED–EM correspondence. Additional fluorescence channels are automatically co-registered with EM ultrastructure, enabling mechanistically meaningful mapping of molecular events. Using HALO-BAK and Drp1, Deep-SegCLEM associates apoptotic membrane discontinuities and mitochondrial fission sites with their ultrastructural counterparts, demonstrating biological relevance. Together, these advances deliver a scalable, end-to-end workflow for quantitative, fiducial-free correlative microscopy, enabling high-throughput mapping of molecular signals to ultrastructure during dynamic cellular processes.

Speaker: Dr Soumaya Zaghbani (Max Planck Institute of Biophysics)

Invited Talks

12 Towards Digital Twins of Nuclear Pores

Speaker: Prof. Martin Beck (Max Planck Institute of Biophysics)

POSTER SESSION

Wednesday 4 March

Invited Talks

13 Using digital twins to probe the origins of intelligence

The origins of perception and cognition have been debated for centuries, with ongoing disagreement about what knowledge and neural structure are hardwired at birth. A major obstacle to resolving this debate is that, until recently, it has been impossible to train formal computational models of intelligence on the same sensory experiences as developing animals. We address this limitation by building digital twins of newborn animals and their environments. We rear newborn animals and “newborn” artificial agents in the same environments, then test the animals and agents on the same tasks. This allows us to provide animals and models with matched learning contexts and test candidate brain models under identical constraints. Using this framework, we show that perceptual and cognitive capacities can emerge from coupling flexible, domain-general neural networks with structured, domain-optimized bodies. Ultimately, this digital-twin framework allows researchers to build formal models of the origins of intelligence, unifying insights across psychology, neuroscience, and artificial intelligence.

Speaker: Prof. Samantha Wood (Indiana University Bloomington)

Short Talk

14 rRNA expansion segments mediate the oligomerization of inactive animal ribosomes

When stressed, all cells downregulate protein synthesis, to conserve energy and shift resources towards repair. Here, we show that in some mammalian cells, including neurons, stress also results in the formation of inactive ribosome-ribosome clusters (disomes). Using cryogenic electron tomography we visualized ribosomes in situ and observed that this dimerization is mediated by a homotypic interaction of rRNA expansion segment ES31Lb. ES31Lb interactions are both necessary and sufficient for disome formation and confer a growth advantage and stress-resistance to cells. ES31Lb is predicted to homo-dimerize in ~20% of chordates, including both a chicken and human variants. Cryo-ET analysis of long-studied chicken tetrasomes revealed an interaction between ES31Lb and ES9La. Taken together, these data indicate a mechanism of translation regulation in animal cells that operates using a flexible component of the protein synthesis machinery - rRNA expansion segments.

Speaker: Ms Mara Müller (MPI for Brain Research)

15 Area-specific signatures of time-irreversibility in spontaneous neural activity of the mouse brain

We study time-irreversibility in spontaneous neural activity as a multiscale dynamical phenomenon. Using large-scale Neuropixels recordings across mouse brain areas, we quantify temporal asymmetry from milliseconds to seconds (30–0.5 Hz) with complementary computational measures capturing population-level asymmetry, metastable state fluxes, and state-space divergence. This framework enables systematic analysis of time-irreversible dynamics across brain regions and timescales.

Speaker: Mr Jonas Elpelt (FIAS)

10:30 Coffee Break

Invited Talks

16 Towards Digital Twins of the cerebral cortex

Speaker: Prof. Sasha van Albada (FZ Jülich)

Short Talk

17 In situ structure of a gap junction - stomatin complex.

Gap junctions (GJs) are intercellular channels that mediate electrical signals and transfer of small molecules. They are crucial for brain, heart, and other organ functions. While molecular structures of purified homomeric GJs are available, information of in situ structures is lacking. In vivo, GJs can form heteromers with different functionalities and may associate with other proteins. Here, we analyzed Caenorhabditis elegans GJs by cryo–electron tomography and subtomogram averaging. We observed hexagonal arrays of GJs at cellular junctions in primary embryonal cells that displayed distinct wide and narrow conformations. Moreover, we found a cap-like, cytosolic protein assembly enclosing the channel pore. We propose that the cap is formed by the stomatin UNC-1, known to interact with UNC-9 innexins. This is corroborated by matching AlphaFold3 models of UNC-1 multimers with our subtomogram average structure; by expressing GFP-tagged UNC-1, leading to cap structures with additional density; and by coarse-grained MD simulations. UNC-1/stomatin rings may affect GJ formation or functions, possibly beyond nematodes.

Speaker: Prof. Alexander Gottschalk (Goethe University)

18 Insights into the plasma membrane association of Extended Synaptotagmin 3

Membrane contact sites between the ER and the PM are site of non-vesicular lipid transport as they harbor various lipid transfer proteins. They are established by protein tethers among which Extended Synaptotagmins (E-Syts) are the most abundant family in mammals. E-Syts are ER-resident and regulate ER-PM connectivity via reversible association of their C2 domains with lipids at the PM. Here, we study exact lipid binding properties of the individual C2 domains of E-Syt3 as well as of a large unstructured linker between the C2B and C2C domains. We use coarse-grained molecular dynamics simulations as well as live-cell confocal and TIRF imaging. We identify novel protein-lipid interactions like PI binding of the C2AB domains and PI4P binding of the C2C domain. Moreover, we show that the unstructured C2B-C2C linker associates with the PM via its aromatic residues. For the C2C domain, we identified two different PM association configurations with different lipid binding profiles.

Speaker: Dr Veronika Thallmair (Philipps University Marburg)

12:30 LUNCH BREAK

Invited Talks

19 Digital Twins in biology and medicine: bridging experimental models, simulation, and health systems

In silico and in vitro technologies are complementary to traditional biological and biomedical tools, enabling the study of multifactorial processes under controlled conditions. In the first part of this talk, I will present examples from bone and joint degeneration and regeneration research, where we combine computer modeling and simulation with microphysiological systems to better understand pathophysiological processes and design regenerative strategies.
In the second part of the talk I will put this work into the context of ongoing work in the European Virtual Human Twins (VHT) initiative, which aims to facilitate the development, credibility assessment, and uptake of digital twins in all areas of health & care, including the basic biological and biomedical research. The recently concluded EDITH Coordination & Support Action brought together the entire VHT ecosystem (including academia, industry, patients, healthcare providers, regulators, payers, policy makers etc) to jointly create a roadmap towards the realization of the VHT. The roadmap (doi:10.5281/zenodo.14769224) provides the context and identified stakeholder needs, followed by a description of the required technology and infrastructure. It continues with a thorough discussion of the relevant standards, regulatory, health technology assessment, legal, ethical and social aspects, as well as the business elements and general incentives. It ends with 30 recommendations that have been formulated for all stakeholders to help advance this moonshot initiative and create a tangible impact on the life of patients.

Speaker: Prof. Liesbet Geris (University of Liège, KU Leuven, VPH society)

Panel discussion: Digital twins - Criteria, Chances and Challenges

20 Digital Twins – Criteria, Chances and Challenges

Panelists: Stefan Knapp, Jan Kosinski,
Loic Royer, Stephan Preibisch,
Liesbet Geris and Sacha van Albada

15:30 Coffee Break

Short Talk

21 Data Driven Modelling of Limb Bud Growth and Morphogenesis

Limb development is regulated by feedback between the tissue-scale mechanics and cellular decisions–migration, contraction, division, and death–implemented through gene-regulatory networks. The dynamically changing states of these cellular networks reflect the decisions being made. To integrate these interactions across 3 scales: genes, cells, and tissues, we develop a 3D cell-based model in which cellular decisions are spatially regulated by morphogen gradients. By constraining parameters with empirical data and optimizing against wild-type limb-bud shape, we obtain a model that links signal distributions to emergent mechanical behaviors. Multiple mechanisms have been proposed for limb-bud morphogenesis, including ectodermal mechanical constraints, proliferation gradients, oriented divisions, distally biased migration, motility gradients, and convergent extension. Yet which spatio-temporal combinations of these activities generate correct morphology remains unresolved. We use the model to test whether convergent extension alone is sufficient to reproduce limb-bud outgrowth and shaping. We compare 2 distinct modes of convergence: (1) orthogonal to an AER/FGF8-associated proximo-distal cue, and (2) parallel to a surface-to-core (radial) cue consistent with Wnt5a. For each mode, we quantify how the strength and timing of convergent extension affects predicted shape trajectories and whether either mode–or a combination of both–can match wild-type morphology.

Speaker: Dr Tim Liebisch (EMBL Barcelona)

22 Initiative of Panoramic Digital Life Model Project

tba

Speaker: Prof. Liangyi Chen (National Biomedical Imaging Center, Peking University)

Invited Talks

23 Building a Virtual Embryo - A Single Cell At a Time.

Imagine having an interactive digital twin of a developing embryo — one you could pause, rewind, or zoom into, exploring how every cell divides, moves, and differentiates. To make this vision real, we created Zebrahub, a dynamic atlas of zebrafish embryogenesis that combines cutting-edge microscopy, powerful computational lineage tracking (Ultrack), and precise molecular mapping into an interactive resource. Our journey begins with advanced multiview light-sheet microscopy, capturing millions of cells in living zebrafish embryos over days of continuous development (DaXi). Next, our uncertainty-aware cell tracker, Ultrack, transforms these enormous terabyte-scale datasets into coherent cellular histories, reconstructing precise lineages even in challenging imaging conditions. Combining these detailed lineages to single-cell transcriptomes, Zebrahub enables users not only to visualize developmental events in unprecedented detail but also to explore the molecular decisions underlying them.

Speaker: Dr Loic Royer (Biohub)

Short Talk

24 Multiscale Investigation of Membrane Remodeling during Selective ER-phagy

FAM134B drives selective ER-phagy to maintain ER homeostasis. Modeling, coarse-grained MD, and experiments capture RHD-driven curvature induction/sensing (isoform-specific), ubiquitination-triggered clustering, and IDR-amplified ER budding. These self-organizing principles inform a hierarchical digital twin of the ER–phagophore contact site linking membrane mechanics/curvature, FAM134B–LC3B tethering, and ATG9-mediated lipid scrambling to predict nascent phagophore recruitment, growth, and maturation.

Speaker: Dr Ramachandra M. Bhaskara (Goethe University)

25 Molecular Digital Twins of Innate Immunity: Insights from Simulations of Human Guanylate-Binding Protein 1

This work explores digital twins for innate immunity, focusing on human guanylate-binding protein 1 (hGBP1) and its role in targeting intracellular pathogens. Using coarse-grained molecular dynamics simulations, we examine protein-membrane interfaces with realistic pathogen membranes. We find that hGBP1 interacts with various membrane types; negatively charged lipids enhance affinity, while lipopolysaccharides are crucial for effective interactions. Additionally, hGBP1 polymerization is essential for stable membrane binding required for pathogen clearance.

Speaker: Prof. Birgit Strodel (FZ Jülich)

CONFERENCE DINNER

Thursday 5 March

Invited Talks

26 Teaching AI the Language of RNA: Foundation Models for Regulation and Therapeutic Design

Computational analysis of high-resolution CLIP-seq data has enabled precise mapping of RBP binding sites and cis-regulatory elements underlying post-transcriptional RNA regulation. Recently, the field has shifted with the emergence of RNA foundation models—self-supervised models trained on vast unlabeled RNA sequences—that enable holistic modeling of RNA function and in silico hypothesis generation. Leveraging such models represents the next frontier, allowing prediction of multiple RNA regulatory processes, as well as in silico design of RNA-based therapeutics. Building on the premise that RNA function is encoded in its interaction partners, we developed Parnet, a multi-task model that densely represents the RNA interactome. Extending our earlier single-task model RBPNet, Parnet is trained end-to-end on raw CLIP-seq profiles from hundreds of RBPs to predict genome-wide binding profiles directly from sequence. In contrast to unsupervised RNA language models, Parnet learns embeddings that capture the combinatorial “RBP code” underlying post-transcriptional regulation. As a result, Parnet generalizes across downstream tasks—including splicing, intron retention, mRNA translation and degradation, and therapeutic design—often with minimal or no fine-tuning.

Speaker: Dr Annalisa Marsico (Computational Health Center, Helmholtz Munich)

27 Mining Molecular Data: from Single Cell Genomics to Cryo-E

Speaker: Dr Judith Zaugg (University of Basel)

10:30 Coffee Break

Short Talk

28 Nuclear speckle periphery organizes stable intron-retained RNAs into a distinct nuclear retention compartment

Intron retention (IR) is an important regulator of RNA fate, yet its spatial and temporal organization in human cells remains poorly understood. Here, we investigate the dynamics, localization, and regulation of intron-retained RNAs in pluripotent stem cells using compartment-resolved transcriptional shutdown, sequence-based modeling, and advanced RNA imaging. We focus on the relationship between IR, nuclear speckles, RNA stability, and cell-cycle–dependent RNA processing.

Speaker: Dr Josep BIayna (Goethe University)

29 From Capability to Confidence: RNA Mass Spectrometry as a Criterion for Digital Twins

Digital twins in structural cell biology critically depend on experimental data quality and interpretability. RNA is a central molecular layer, yet its mass spectrometric analysis remains analytically fragile. I will discuss why RNA-MS cannot be treated as an extension of proteomics and how intrinsic ambiguity, redundancy, and modification complexity challenge confidence. I argue that explicit analytical control and RNA-aware validation criteria are essential for integrating RNA into mechanistically meaningful digital twins.

Speaker: Prof. Stefanie Kaiser (Goethe University)

Invited Talks

30 Digital twins for modelling gene expression: applications to association studies and personalized drug recommendation

Modern applications of AI technologies have enabled to predict the expression of genes genome-wide in a cell-type specific way using DNA sequence or epigenetic information. In this talk I will explain how these methods can be used to build digital twins for modelling gene expression and introduce novel applications for associating genes with disease and suggest personalised drug treatments.

Speaker: Prof. Marcel Schulz (Goethe University and Uniklinikum Frankfurt)

12:30 Lunch Break

Short Talk

31 Molecular mechanism of membrane pore formation triggered by PI(4,5)P2- dependent FGF2 oligomerization

Fibroblast Growth Factor 2 (FGF2) is a key cell survival factor involved in tumor-induced angiogenesis. Unlike most secreted proteins, FGF2 lacks a signal peptide and is exported via unconventional protein secretion (UPS), bypassing the ER/Golgi. It translocates directly across the plasma membrane (Type I UPS), a process initiated by PI(4,5)P₂- mediated recruitment to the inner leaflet. Subsequently, FGF2 forms membrane-spanning oligomers within toroidal membrane pores. The final stage of secretion involves heparan sulfate proteoglycans on the cell surface, which disassemble these oligomers at the outer membrane leaflet, ensuring FGF2 is released into the extracellular space to mediate autocrine and paracrine signaling. Using atomic force microscopy, biochemical reconstitution experiments, and multiscale molecular physics-based computer simulations, we show that FGF2 self-assembles into a ring-like structure, triggering PI(4,5)P₂ lipid sorting and membrane remodeling. When PI(4,5)P₂ reaches a critical threshold, its non-bilayer properties destabilize the membrane, forming a pore through which FGF2 translocate across the membrane. We propose that the geometry of FGF2 oligomers and their lipid sorting capacity are crucial for protein translocation, suggesting a shared mechanism with other pore-forming proteins that assemble into ring-like structures

Speaker: Dr Fabio Lolicato (Heidelberg University)

32 Learning Biomolecular Ensembles from Experimental Cryo-EM Data

Biomolecules are inherently flexible, continuously transitioning between conformations to carry out their cellular functions. Cryo–electron microscopy (cryo-EM) enables us to image individual biomolecules at near-atomic resolution. As a result, an image dataset can capture the entire biomolecular ensemble at once. By comparing cryo-EM images to structural hypotheses, one can identify the structure that best matches each image. In principle, this allows us to identify the structures in the images and infer the underlying ensemble. However, performing these exhaustive comparisons is computationally very expensive. To tackle this problem, we introduce simulation-based inference for Cryo-EM (CryoSBI), a novel method that utilizes simulation-based inference to infer biomolecular conformation from individual cryo-EM images. Our approach uses neural posterior estimation, a technique that directly approximates the Bayesian posterior using a simulator and a density estimator. Training is performed only once using simulated data generated with the simulator. Afterward, inference for each particle requires just a single forward pass through the neural network. This eliminates the need to estimate particle pose and imaging parameters for each observation, delivering substantial computational speedups compared to traditional explicit likelihood methods. We demonstrate this approach through experiments on real cryo-EM data, where cryoSBI successfully extracts molecular conformations with reliable confidence measures. In addition, we combine cryoSBI with established ensemble reweighting techniques to infer biomolecular ensembles directly from entire cryo-EM image datasets. This enables the recovery of population distributions over conformational states. We demonstrate this capability by reweighing p4p6-conformational ensembles using cryo-EM data collected under different environmental conditions, revealing how changes in the experimental environment modify the underlying ensemble.

Speaker: Mr Lars Dingeldein (FIAS)

Invited Talks

33 Insights into the mitochondrial collective

Speaker: Prof. Suliana Manley (EPFL - Laboratory of Experimental Biophysics Lausanne)

15:30 Coffee Break

Short Talk

34 The role of SAM domain in ΔNp63α

he p53 family member p63 exhibits multiple isoforms and functional domains, with ΔNp63α serving as a critical transcriptional regulator implicated in epithelial-mesenchymal transition. While the various domains of ΔNp63α have been extensively characterized, the precise function of its Sterile Alpha Motif (SAM) domain stays elusive. Mutations within the SAM domain are associated with Ankyloblepharon-Ectodermal Defects-Cleft Lip/Palate (AEC) syndrome and Rapp-Hodgkin syndrome (RHS), primarily due to SAM domain destabilization. These conditions are characterized by severe skin erosions, underscoring the domain's significance in maintaining epithelial integrity. SAM domains are ubiquitous in eukaryotic proteins and are renowned for their propensity to oligomerize through diverse mechanisms. However, the p63 SAM domain has not previously demonstrated such oligomerization tendencies. Intriguingly, the presence of a putative zinc-binding motif within the SAM domain prompted our investigation into zinc-mediated effects using Nuclear Magnetic Resonance (NMR) spectroscopy. In this study, we present evidence of zinc-induced dimerization of the p63 SAM domain and elucidate its dimeric structure. Additionally, we explore the potential implications this dimerization might have on the full protein.

Speaker: Ms Srilakshmi Kalathil (Goethe University)

35 Parameterization of Molecular Photoswitches for Martini 3 and their Application in Photopharmacology

Molecular photoswitches are light-responsive compounds that undergo reversible conformational changes upon irradiation. We present Martini 3 coarse-grained models for seven photoswitches, parameterized using semi-empirical reference data and experimental water–octanol partitioning. These models enable the development of digital twins for light-controlled biological systems including membrane modulation, peptide conformational control, and ion channel regulation to study photopharmacology at system sizes and time scales inaccessible to atomistic simulations.

Speaker: Mr Thilo Duve (FIAS)

Invited Talks

36 If Machines Can Learn, Who Needs Scientists?

Machine learning is impacting not just the natural sciences but also the social sciences, engineering, architecture, and the art world. In many fields an obstacle to the application of machine learning is the relative paucity of available training data. Other challenges
include the problem of interpreting the results of a machine learning algorithms, incorporating machine learning into hypothesis-driven research, and ethical and reproducible use. This perspective examines the potential of machine learning in NMR structural biology, the role of scientists, and speculates on possible approaches to the hurdles.

Speaker: Prof. Jeffery Hoch (UConn Health)

37 In-cell structural systems modeling

One of the central goals of computational biology is to build a realistic model of a cell – one that can simulate cellular processes, predict the effects of perturbations, and support rational drug design. This vision has gained new momentum through recent technological advances, for example, in single-cell omics, structural proteomics, and visual mapping of cells using in-cell cryo-electron tomography (cryo-ET) and volume electron microscopy. The emergence of AI-based methods such as AlphaFold and other AI-driven models of biological complexity further contributes to this renewed interest. Although the form of such a comprehensive cell model remains undefined, the recent technological advances now allow us to build components that were previously out of reach. In this context, I will present our contribution toward modeling a cell by leveraging some of these recent technological advances. I will show the computational tools we are developing as part of an effort to build a computational framework for in-cell structural systems biology, aimed at enabling spatial and systems-level mapping of the cellular environment. These will include methods for data-driven modeling macromolecular assemblies and interaction networks, as well as approaches for identifying and modeling molecules and membranes in cryo-ET data.

Speaker: Mr Jan Kosinski (EMBL)

Closing Remarks