Molecular biophysics at the transition state: From statistical mechanics to AI

7 Jul 2025, 14:00 → 11 Jul 2025, 12:45 Europe/Rome

Palazzo Consolati - University of Trento

Alessandro Ingrosso, Francesca Mele (University of Trento), Roberto Menichetti (Physics Department, University of Trento), Lorenzo Petrolli (University of Trento), Raffaello Potestio (UniTn - Physics Dept.)

Welcome to the Trento edition of the StatPhys29 satellite workshops! As the title suggest, the event will gather young researchers and leading experts in the fields of computational biophysics, statistical mechanics and machine learning, to discuss the challenges and crossroads that stand before biomolecular simulations in the machine learning era.

Here (and in the dedicated sections of the website) you’ll find all relevant information on the workshop and the venue - yet, feel free to contact us anytime at biomol.ai_ws.phys@unitn.it.

Posters and oral contributions will be admitted upon evaluation and acceptance, and participation in the event will be totally free of charge. We look forward to seeing you in Trento!

CALL FOR ABSTRACT DEADLINE EXTENDED - New deadline: May 25th 2025

---

Purpose of the workshop

The comprehension of the intimate behaviour of biological macromolecules and molecular assemblies has absorbed the decades-long efforts of statistical physicists. Indeed, the framework of equilibrium, and more recently non-equilibrium statistical mechanics has provided the main strategy to address fundamental questions on the biological processes that take place at the microscale: these range from protein folding to catalysis, from allostery to signal transduction, from RNA transcription to gene expression, just to name a few. The ubiquitous presence of thermal fluctuations requires the platform of statistical mechanics to solve such problems, and theoretical and/or computational methods have been developed to this end.

 

Recently, though, a new player has entered the game, that is machine learning. The ever-growing variety of methods and techniques that falls under this term is gaining ground in the toolbox of molecular biophysicists, and with the most diverse scopes: ML can be employed to speed up calculations, to carry out standard analyses in a faster manner, to provide new perspectives on old problems, or even to generate knowledge - from molecular structures to thermodynamic functions.

 

The field is at a critical crossover, in which theory-based approaches are flanked by novel tools, which come with a threatening aura and the promise of greatness. In this workshop, we will gather world-leading experts in several fields, from applied statistical mechanics to computational biophysics and machine learning science, to understand the challenges and opportunities that lie at the interface between the most successful established methods and novel, AI-based techniques.

---

Confirmed Invited Speakers

• Paolo Carloni (Forschungszentrum Jülich, Germany)

• Michele Cascella (University of Oslo, Norway)

• Oleksandra Kukharenko (Max Planck Institute for Polymer Research, Germany)

• Rémi Monasson (CNRS & Ecole Normale Supérieure, France)

• Agnes Noy (University of York, United Kingdom)

• Elena Papaleo (Technical University of Denmark, Denmark)
• Jan Schuemann (Massachusetts General Hospital & Harvard Medical School, USA)
• Giuseppe Jurman (Fondazione Bruno Kessler & Humanitas University) 
  

 

---

Important dates and information

• Registration and abstract submission opening: February 15th 2025
  

• (New) Abstract submission deadline: May 25th 2025
• Registration deadline: May 30th 2025
   
• To apply for an Oral or Poster contribution, please refer to the Call for Abstracts section of the website and fill in the associated form.
• To register for the workshop, please refer to the Registration section of the website and fill in the associated form.
• Please note that both abstract submission and registration for the event require you to have or create an Indico account. Should you need support, do not hesitate to contact us or the ECT* staff.

 

---

Social events

Two social events are foreseen during the workshop, namely:

• A welcome networking aperitivo + food (Monday, July 7^th, 18:30), which will be hosted by La Bookique. Expenses for this event will be covered by the organization.

• The social dinner (Wednesday, July 9^th, 20:00), which will take place at Trattoria Piedicastello. Please note that expenses for this event (about 30 €) will be covered by the participants.

 

---

WARNING: for hotel bookings do not give your data to any external services upon email request; there are known cases of fraud. A list of accommodations, together with the instructions to obtain a discount rates (limited option, available upon "early-bird" registrations) can be found here.

---

Co-organizing Institutions

---

Sponsors

---

 

Monday 7 July

14:00 → 14:30 Welcome and introductory lecture (Raffaello Potestio, Alessandro Ingrosso)

14:30 → 15:20 Mechanisms of self-assembling of non-conventional surfactants by multiscale modelling and data-driven approaches

The dynamics regulating self-organisation of surface-active compounds is
at the basis of the living matter; for example, it is involved into the definition of cellular boundaries, cell compartmentalisation and molecular trafficking. Surfactants are also broadly used in soft matter technology, from responsive materials, to nanodelivery. The problem associated with their characterisation at molecular level lies in the non-reducible size of the systems of interest, spanning several orders of magnitude from the nm to the $\mu$m, and in their relatively long relaxation times, which can often pass the millisecond. Here, I will give an overview of recent advances in modelling of the self-assembling dynamics of surfactants, using a hierarchy of approaches at different resolution. I will discuss how external factors like ionic-strength, or photoactivation can have a major role in controlling aggregation [1].
I will also show how combining soft density-functional based models [2] to enhanced sampling metainference approaches [3] can be used to describe the aggregation of non-conventional surfactants that do not respect the common core-shell packing [4], and propose data-driven approaches for systematic calibration of such models [5].

References

1. (a) K. Schäfer et al. Angew. Chem. Int. Ed. 59, 18591-18598 (2020); (b) V. A. Bjørnestad et al. J. Colloid Interface Sci. 646, 883-899 (2023)
2. M. Ledum et al J. Chem. Theory Comput. 19, 2939-2952 (2023)
3. (a) M. Bonomi et al. Sci. Adv. 2, e150117 (2016); (b) H. M. Cezar, M. Cascella J. Chem. Inf. Model. 63, 4979-4985 (2023)
4. H. M. Cezar et al. Small Struct. 2400553 (2024)
5. M. Carrer et al. J. Chem. Inf. Model. 64, 5510-5520 (2024)

Speaker: Michele Cascella (University of Oslo)

15:20 → 15:50 Advancing the Prediction of Binding Events in Highly Flexible, Allosteric and Multidomain Proteins

Accurately predicting ligand-protein interactions remains a cornerstone of rational drug discovery [1,2]. Traditional docking methods often struggle to capture the dynamic conformational landscapes of such proteins, especially when only apo (unbound) structures are available [1]. In this work, I will introduce our recent protocol gEDES (generalized Ensemble Docking with Enhanced sampling of pocket Shape) [2,3], a computational method designed to generate holo-like conformations from apo protein structures. gEDES employs enhanced sampling techniques to explore the conformational space of proteins, focusing on the dynamic reshaping of binding pockets and sub-pockets that are critical for ligand binding. This approach enables the modeling of induced-fit effects and allosteric transitions without prior knowledge of the bound state. We applied gEDES to adenylate kinase, a prototypical allosteric enzyme with significant domain movements upon ligand binding. Our results demonstrate that gEDES accurately reproduces holo-like conformations, facilitating precise docking of substrates, inhibitors, and non-competent analogs. Compared to state-of-the-art deep learning methods such as NeuralPLexer [4], which utilizes multiscale generative diffusion models for protein-ligand complex prediction, gEDES exhibits superior sensitivity to subtle chemical variations in ligands, leading to more accurate binding pose predictions [2]. When benchmarked against the DUD-E (Directory of Useful Decoys, Enhanced) database, a comprehensive resource for evaluating docking and virtual screening performance, gEDES was able to generate druggable conformations for more than 80% of the targets. gEDES will be made available to the through a user-friendly web server to make gEDES accessible to the broader scientific community, facilitating the generation of holo-like structures and the setup of simulations for diverse protein targets.

Speaker: Han Kurt (University of Cagliari)

15:50 → 16:20 Interaction of SARS-CoV-2 Fusion Peptides with Lipid Monolayers by Molecular Dynamics Simulations

Membrane fusion between the viral envelope and host cell is a key stage during the infectivity of SARS-CoV-2 that enables the release of the viral genetic material into the target cell. Within the region of the Spike protein implicated in this process, several peptides have been identified to interact with the cell membrane [1]. Through Molecular Dynamics (MD) simulations, we aim to elucidate the role of these fusion peptides in viral infection by investigating their interactions with biomimetic model membranes, using lipid monolayers to represent a membrane leaflet. We first use all-atom (AA) MD simulations to obtain a stable conformation for each peptide in physiological conditions. The resulting structures are then mapped into a coarse-grained (CG) model in order to increase the length and timescales for the studied systems. The CG model we employ, Martini 2, developed to simulate the behavior of a lipid membranes [2] and proteins [3], has been proven to be transferable to a wide range of biological systems [4]. We explore the effect that each peptide has on the interaction with biomimetic lipid monolayers under varying surface area as well as how lipid composition affects their binding to the membrane.

References
[1] A. Santamaria et al., JACS 144, 2968-2979 (2022)
[2] S. J. Marrink et al., J Phys Chem B 111, 7812-7824 (2007)
[3] L. Monticelli et al., J Chem Theory and Comput 4, 819-834 (2008)
[4] S. J. Marrink and D. P. Tieleman, Chem Soc Rev 42, 6801-6822 (2013)

Speaker: Sebastian Jimenez Millan (Centro de Física de Materiales (CFM-MPC), CSIC-UPV/EHU, Donostia-San Sebastián, Spain)

Sebastian_Jimenez_Millan.pdf

16:20 → 17:05 Coffee break

17:05 → 17:35 Genome replication dynamics in E. coli and Yeast and their effect on cellular growth

To allow cells to proliferate successfully and efficiently, genome replication is governed by a sophisticated replication program which governs the time at which different parts of the genome are replicated. These programs are carefully synchronized with the cell cycle to ensure timely replication without undue interference with other cellular processes. We quantitatively study the replication program and its interactions with the cell cycle in both E. coli and yeast using replication timing data. Replication timing data are DNA abundance profiles measured in exponentially growing, asynchronous populations from which we infer a wide range of properties of the underlying replication program [1]. In E. coli a machine learning approach based on Gaussian Processes allows us to measure the temporal fluctuations of replication fork velocity in both wildtype and ectopic origin mutants with high precision. We observe that additional forks reduce population growth and fork velocity, which suggests mutual competition between forks and between forks and other cellular processes. Based on our observations, we introduce a model of resource allocation between replication and cellular growth and show that it is consistent with our data. In yeast we demonstrate that our method can infer the location and efficiency of origins of replication from replication timing data without any prior information except the (unannotated) genome. Since our method only requires organisms to be cultivable in the lab, this opens the door to studying replication programs and a wide range of single-cellular species.

[1] F. Pflug, D. Bhat, S. Pigolotti, PLOS Computational Biology 20(1):e1011753, (2024).

Speaker: Florian Pflug (Okinawa Institute of Science and Technology)

Florian_Pflug.pdf

17:35 → 18:05 Cooperativity in Transcriptional Regulation: A Molecular Dynamics Approach

Objectives

Transcription factors (TFs) are the primary drivers of gene regulation by binding to specific DNA sites with adequate affinity and stability [1]. TFs often cooperate as homodimers or heterodimers for a more delicate regulation [2]. Granyhead-like 3 (GRHL3) and Hepatocyte nuclear factor 4 alpha (HNF4α) are TFs that form heterodimers to initiate mesenchymal to epithelial transition (MET), a cell-fate alteration important in development and cancer progression [3]. In this work, we investigate the role of DNA sequence in modulating the cooperativity of transcription factors.

Methods

We developed atomistic structural models of free or nucleosomal DNA bound to human GRHL3, as well as models also including HNF4α, and performed molecular dynamics simulations.

Results

Our molecular dynamics simulations show that the two TFs cooperate when GRHL3 is bound to its native DNA sequence, but the cooperativity is disrupted by mutations in the specific DNA sequence, potentially halting MET progression.

We also investigated NF-κB p50, a TF involved in inflammation, immune response, cell division, cellular differentiation, and survival. Simulations of DNA-bound NF-κB p50 revealed homodimeric cooperativity and progressive DNA bending—suggestive of transcription-associated DNA melting. We find that the TF specificity can be established by not only the cognate DNA sequence but also by the flanking sequences upstream and downstream.

Conclusions

Our findings demonstrate that the DNA sequence is crucial for the heterodimer or homodimer cooperativity between TFs, consequently for the DNA dynamics and conformation, and gene regulation.

Acknowledgements

This project is supported by the Tübitak 2250 awarded to G.Ç., EMBO Installation Grant No:5056 and EuroHPC MareNostrum5 No:REG-2023R02-125 awarded to S.K., Tübitak 2247-A No:120C149 and Tübitak 1001 No:122Z215 awarded to D.A.

References

[1] Aykut Erbas and John F. Marko, Curr Opin Chem Biol., 53, 118–124 (2019).

[2] Jacqueline M. Matthews. Protein Dimerization and Oligomerization in Biology. Springer Science & Business Media (2012).

[3] Burcu Sengez et al., Cells, 8, 858 (2019).

Speaker: Gözdem Çavdar (Izmir Biomedicine and Genome Center & Dokuz Eylül University)

Gozdem Cavdar

Tuesday 8 July

09:00 → 09:50 A structure-based framework to investigate variant effects in cancer and other diseases

The growing volume of genomic data has led to a surge in missense variants. However, many of these remain classified as Variants of Uncertain Significance (VUS) or have conflicting annotations, limiting their utility in diagnostics and precision medicine. Predictive tools often lack resolution into how such variants affect protein function at the molecular level. To address this, we present MAVISp (Multi-layered Assessment of VarIants by Structure for proteins), a modular framework that integrates structural modeling, free energy calculations, and interaction analyses to characterize the mechanistic impact of protein-coding variants. Available as a publicly accessible web resource (https://services.healthtech.dtu.dk/services/MAVISp-1.0/), MAVISp includes curated data for over 700 proteins and more than five million variants, with ongoing updates by a team of expert biocurators. We will showcase its utility through selected case studies, illustrating how MAVISp enables the interpretation of structural stability changes, disruption of protein–protein and protein–DNA interactions, and perturbations of post-translational regulation. In particular, we will present examples of applications to proteins involved in DNA repair, DNA damage response, and autophagy, highlighting how structural insights can refine variant prioritization and support functional reclassification. This integrative strategy bridges the gap between sequence-based prediction and protein-level mechanisms, facilitating the interpretation of mutational landscapes in cancer and beyond.

Speaker: Elena Papaleo (Technical University of Denmark)

09:50 → 10:20 Evolutionary constraints guide AlphaFold2 in predicting alternative conformations and inform rational mutation design

Investigating structural variability is essential to understand protein biological functions. Although AlphaFold2 accurately predicts static structures, it fails to capture the full spectrum of functional states. Recent methods have used AlphaFold2 to generate diverse structural ensembles, but they offer limited interpretability and overlook the evolutionary signals underlying predictions. In this work, we enhance the generation of conformational ensembles and identify sequence patterns that influence alternative fold predictions for several protein families. Building on prior research that clustered Multiple Sequence Alignments to predict fold-switching states, we introduce a refined clustering strategy that integrates protein language model representations with hierarchical clustering, overcoming limitations of density-based methods. Our strategy effectively identifies high-confidence alternative conformations and generates abundant sequence ensembles, providing a robust framework for applying Direct Coupling Analysis (DCA). Through DCA, we uncover key coevolutionary signals within the clustered alignments, leveraging them to design mutations that stabilize specific conformations, which we validate using alchemical free energy calculations from molecular dynamics. Notably, our method extends beyond fold-switching, effectively capturing a variety of conformational changes.

Speaker: Valerio Piomponi (Scuola Internazionale Superiore di Studi Avanzati (SISSA))

10:20 → 11:05 Coffee break

11:05 → 11:35 Toward Reliable Synthetic Omics: Statistical Distances for Generative Models Evaluation

Synthetic data generation is emerging as an approach to overcome the limitations of real-world data scarcity in omics studies, especially in precision medicine and oncology. Omics datasets, with their high dimensionality and relatively small sample sizes, often lead to overfitting, especially in deep learning models. Generative models offer a promising way to generate realistic synthetic data preserving the original data distribution. However, there is still no objective consensus on how to evaluate their performance. In this talk, we set out to validate generative networks for transcriptomics data generation by using statistical distances as robust evaluation metrics. In particular, we observe that statistical distances enable simultaneous evaluation of global and local data fidelity of generated synthetic data. Because these distances satisfy the properties of true metrics, they also enable formal hypothesis testing to assess whether generative models have in fact converged or are merely approaching the reference distribution. Crucially, optimizing for these distances was found to implicitly select models maximizing other widely used metrics of generative performance, providing evidence of their broad applicability. Overall, our findings indicate that the adoption of these metrics can play a key role in guiding the development of generative models across a wide range of domains.

Speaker: Giuseppe Jurman (Fondazione Bruno Kessler & Humanitas University)

11:35 → 12:05 S100B inhibits the formation of Aβ42 fibrils and intermediate oligomers implicated in Alzheimer´s disease

Alzheimer’s disease (AD) neurodegeneration involves aggregation of the amyloid beta-42 peptide (Aβ42) into transient oligomers and fibrils, whose emergence is regulated by a limited set of extracellular chaperones. Among these is S100B, a homodimeric protein that is up-regulated in AD and which acts as a Ca2+-activated Aβ42 amyloid suppressor. Nonetheless, S100B occurs in the human brain also as a homotetramer (Ostendorp T et al., 2007 EMBO J.), whose AD-linked neuroprotective functions remain uncharacterized.

Here we present recent research in which we establish and compare the Aβ42 anti-aggregation and anti-oligomerization activities of both S100B multimers. Using thioflavin-T monitored Aβ42 aggregation kinetics, we discovered that unlike the dimer, tetrameric S100B inhibits Aβ42 aggregation even in the absence of Ca2+ binding, while operating at sub/equimolar ratios. Next, we used computational predictors of aggregation-prone regions to map surfaces within tetrameric S100B amenable to interact with Aβ42. We found a secondary Ca²⁺-independent cleft that facilitates binding to both Aβ42 monomers and fibrils, as corroborated by circular dichroism, electron microscopy and docking calculations (Figueira AJ et al., 2022 J. Mol. Biol.). Our investigation additionally explored the impact of such S100B multimers on the generation of Aβ42 intermediate oligomers (AβO). For this, we fitted Aβ42 aggregation traces to mathematical models describing the mechanisms of Aβ42 fibrillation (Meisl G et al., 2016 Nat. Protoc). This revealed that dimeric and tetrameric S100B inhibit Aβ42 nucleation catalysed by fibril surfaces, decreasing the reactive influx towards oligomers down to <10% and reducing the total amounts of AβO by 30-60% (Figueira AJ et al., 2023 Front. Neurosci.).

Taken together, our findings highlight S100B multimers as versatile and complementary inhibitors of Aβ42 neurotoxic oligomerization and aggregation, hinting their pivotal role in the regulation of AD synaptic proteostasis.

Funded by EU-TWIN2PIPSA/GA101079147 and FCT-Portugal BD/06393/2021 (AJF)/UID/MULTI/04046/2020 (BioISI).

Speaker: António J. Figueira (BioISI – Instituto de Biosistemas e Ciências Integrativas, Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisboa, Portugal | Departamento de Química e Bioquímica, Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisboa, Portugal)

Oral_contribution_AFigueira.pptx

12:05 → 12:35 The effects of knot topology on the collapse of active polymers

We use numerical simulations to study tangentially active flexible ring polymers with different knot topologies. Simple, unknotted active rings display a collapse transition upon increasing the degree of polymerization. We find that topology has a significant effect on the polymer size at which the collapse takes place, with twist knots collapsing earlier than torus knots. We rationalize this behavior as a consequence of the propensity for non-neighboring bonds of torus knots to be aligned with each other, thus avoiding collisions that would eventually lead to a MIPS-like collapse.

Speaker: Davide Breoni (Università di Trento)

12:35 → 14:30 Lunch break

14:30 → 15:20 Generative model of SARS-CoV-2 variants under immune pressure unveils viral escape potential

The evolutionary trajectory of SARS-CoV-2 is governed by competing pressures for ACE2 binding, structural viability and escape from neutralizing antibodies targeting its receptor-binding domain (RBD). Here, we present a modular framework that quantifies immune selection and predicts antibody resilience beyond single mutations or known variants, by integrating deep mutational scanning (DMS) measurements of ACE2 affinity and escape profiles for 31 monoclonals with a generative sequence model trained on pre-pandemic Coronaviridae. To assess the escape potential of individual antibodies, we designed RBD variants under pressure from four clinically relevant antibodies (SA55, S2E12, S309, VIR-7229). Of 22 tested designs, bearing up to 21 mutations from Wuhan, 50% expressed as stable protein. Binding assays confirm that S309 and VIR-7229 retain recognition across diverse mutation combinations. DMS-informed mutational effects confered strong predictive power, successfully forecasting which antibodies can be subverted by our designed backgrounds. Finally, by identifying negatively correlated escape routes, we prioritize antibody combinations with minimal shared vulnerabilities. By quantitatively linking viral adaptation to antibody resistance profiles, this model provides a predictive foundation for optimizing therapeutic strategies and enhancing long-term pandemic preparedness.

Speaker: Remi Monasson

15:20 → 15:50 MADRNA: Coarse Graining RNA Force Fields via Machine Learning

In Protein structure prediction there have been massive improvements recently with the help of machine learning. In RNA structure prediction however the situation is less ideal due too much sparser experimental data. In principle, molecular dynamics (MD) can provide access to structural dynamics, but the relevant time scales are often out of reach due to prohibitive computational expenses.

A promising approach to scale MD to larger molecules and time scales is machine learning (ML)-based coarse graining.
It combines (i) systematic coarse graining, where the degrees of freedom (DOFs) of a molecular system are systematically reduced, and (ii) ML potentials that model the interaction of the new coarse-grained degrees of freedom.

Our approach is based on variational force matching, i.e., we generate trajectories of RNA using classical all-atom simulations, project them down to a coarse-grained level, and train a ML potential to minimize the discrepancy between predicted forces and the forces obtained in simulations.
Due to crucial stabilizing effects such as base pairing and stacking, special care is needed when specifying the coarse graining.
Our ML potential is built upon the graph neural network SchNet, which enables learning the many-body interactions that govern RNA dynamics.

Preliminary results on a set of tetraloops indicate that our method is able to capture the behavior of RNA correctly. At the same time, it reduces the number of simulated atoms by 2 orders of magnitude and allows for a step size twice as large during simulation, compared to the all-atom setting.

Speaker: Anton Dorn (Forschungszentrum Jülich)

CGModels_animated.pdf

15:50 → 16:20 Exploring the role of protein folding intermediates: cryptic phosphosites as a new mechanisms of protein homeostasis regulation

As the name suggests Post-Translational Modifications (PTMs) are commonly believed to happen after the production of the target protein is fully completed. However, by examining residues involved in the most widespread PTM, namely phosphorylation, one finds that some such residues cannot be accessed when the protein has reached its functional form. Indeed, one finds that a part of the experimentally validated phosphosites are buried inside the hydrophobic core of proteins in the native state.
We evaluated the extent of this phenomenon in the case of the entire human proteome by computing the relative solvent accessibility (rSA) of phosphorylation sites in the 3D structures predicted by AlphaFold2. To generate a more conservative dataset we then applied a filtering procedure based on the division of proteins into quasi-rigid domains. What we found is that roughly 5% of all phosphorylation sites are cryptic, which translates to about one out of three phosphoproteins having at least one cryptic phosphosite.
We hypothesize that these cryptic phosphosites may be exposed along the folding pathway and serve as a previously unappreciated mechanism for protein quality control.

Speaker: Annarita Zanon (University of Trento)

16:20 → 17:05 Coffee break

17:05 → 18:30 Poster session

Wednesday 9 July

09:00 → 09:50 Investigating membrane proteins by QM/MM MD and machine learning in the Exascale era

First principle quantum mechanical/molecular mechanics (QM/MM) molecular dynamics (MD) simulations constitute an excellent methodology for the study of quantum phenomena in biological systems. Here we will present QM/MM MD studies from our lab on receptors and transporters. These studies demonstrate the expanding domain of applicability of the approach, thanks to the advent of large parallel machines (such as the exascale computers) and ML.

Speaker: Paolo Carloni (Forschungszentrum Jülich GmbH)

09:50 → 10:20 Enhancing Molecular Simulations by Integrating Machine Learning and Quantum Computing

The rare event problem often limits practical applications of classical Molecular Dynamics simulations. While powerful enhanced sampling techniques have been proposed and successfully applied to many important study cases, these methods typically require a suitable choice of collective variables (CVs) in input. This feature introduces systematic errors that are hard to estimate a priori. In this talk, we shall discuss a novel approach to enhanced sampling that does not require any CV in input. Instead, it relies on a combination of generative AI and uncharted exploration to enhance sampling in a fully data-driven way and to renormalization group techniques to use the generated data to represent equilibrium dynamics. In the resulting theory, path sampling can be further enhanced by resorting to quantum annealing machines, which enable the storage of the entire transition path ensemble in a single quantum computer's wave function.

Speaker: Pietro Faccioli (University of Milan-Bicocca)

10:20 → 11:05 Coffee break

11:05 → 11:35 RNA secondary structures scale as self-avoiding randomly branched polymers

Ribonucleic acid (RNA) is a heteropolymer of four nucleotides (A, C, G, and U) which interact and form different base pairs with each other. Base pairing bewteen nucleotides in an RNA sequence gives rise to a secondary structure where base-paired (double-stranded) helices are interspersed with stretches of unpaired (single-stranded) nucleotides. In long RNA sequences, this leads to formation of complex structures with branching architectures, whose topology often comes with functional consequences such as increased compactness or enhanced interactions with the environment. When RNA secondary structures are mapped to mathematical trees, the structures and their topology can be analyzed within the framework of the statistical physics of branched polymers. I will show that a Flory-type analysis applied to the topological properties of RNA structure results in characteristic scaling exponents that are similar to those of self-avoiding randomly branched polymers in two and three dimensions. This result is surprising as the prediction of RNA secondary structure depends solely on the nearest-neighbour energies of base pair formation and does not include any steric interactions, which otherwise distinguish self-avoiding and ideal polymers. Furthermore, I will demonstrate that the scaling behaviour of RNA structures is robust across different sequence compositions and energy models, persisting down to the most bare-bones models of secondary structure. Only if the actual tree topologies are shuffled does one obtain the scaling exponents characteristic of ideal branched polymers. Put differently, the topologies of RNA secondary structures consistently assume only a specific subset of all possible tree topologies in a way that makes their scaling behaviour similar to the one of self-avoiding branched polymers. Our work explores the conditions under which this remains true as well as in what way the scaling properties are encoded in the sequence and structure of RNA.

[1] D. Vaupotič et al., arXiv preprint, arXiv:2409.16007 [cond-mat.soft] (2024).
[2] D. Vaupotič et al., J. Chem. Phys. 158, 234901 (2023).

Speaker: Anze Bozic (Department of Theoretical Physics, Jozef Stefan Institute)

ECT2025_ABozic.pdf

11:35 → 12:05 Topological polymeric soft materials

Materials built of topologically interlocked polymer rings have recently gained considerable interest in supramolecular chemistry, biology, and soft matter. Two typical exaples are polycatenanes, linear chains of concatenated rings, and the kinetoplast DNA (kDNA), the mitochondrial genome of trypanosomatids, formed by ~5000 dsDNA minirings linked together to form a 2D surface whose topology is on average conserved through replication. Here I present the results of several ongoing collaborative efforts, all highlighting the role of topological interactions in shaping the physical properties of supramolecular objects and how one can exploit them to tune the behavior of bioinspired materials. I will show that circular polycatenanes can topologically trap twist and behave similarly to supercoiled dsDNA, and that a similar effect holds for 2D sheets of rings. Finally, I will show how coarse-grained (CG) simulations of kDNA can be used together with experimental data to clarify its properties. Our results suggest that supramolecular topological objects can form a new category of highly designable structures with potential applications in supramolecular chemistry and material science.

Speaker: Luca Tubiana (University of Trento, Italy)

12:05 → 12:35 Interactive MD in Virtual Reality to Explore Macromolecular Landscapes

Molecular systems show intricate dynamics due to many particles' interactions, making them challenging to study even for specialized researchers. Molecular dynamics (MD) simulations have proven to be an effective tool to investigate the kinetics of biological processes. However, simulations often require extensive sampling or computational resources to gather data on rare events, which are often the focus of scientific investigation. Methods like Umbrella sampling, metadynamics, and string methods accelerate the analyses but require a priori information to describe the investigated process, such as a reaction coordinate (RC) or collective variable (CV) subspace, which may not be obvious to define in advance.
Interactive molecular dynamics (iMD) is a framework enabling real-time manipulation of MD simulations. Specifically, iMD in Virtual Reality (iMD-VR) offers this feature in an immersive 3D environment, allowing a direct and intuitive approach to interacting with molecular systems [1]. It has been shown to have the potential to combine human reasoning and molecular design insight with computational automation and has previously been used to accelerate rare events in protein-ligand binding through user-guided manipulations [2][3].
An interesting aspect of this approach is that there is no need to define a set of CVs a priori. This allows the researcher to freely explore the systems' conformational space, which is extremely useful when the RC is unclear.
NanoVer is a free, open-source software that facilitates iMD-VR, delivering quantitative information about the system and user interaction (energy, forces, cumulative work done, etc.). Based on this data, users can analyze qualitative preliminary profiles of the conformational landscape, assessing how the system behaves along paths that bias it from state A to state B. User-guided paths can then be extracted and used as RCs to obtain free energy profiles with other established MD techniques [4]. This human-in-the-loop workflow combines the chemical intuition of the researcher with nonequilibrium-enhanced sampling techniques, accelerating the investigation of macromolecular systems' kinetics.

References:
[1] O’Connor et al., Sci. Adv., 4, eaat2731 (2018)
[2] O’Connor et al. J. Chem. Phys., 150, 220901 (2019)
[3] Shannon et al. J. Chem. Phys., 155, 154106 (2021)
[4] Deeks et al. Sci Rep, 13, 16665 (2023)

Speaker: Ludovica Aisa (University of Santiago De Compostela USC)

Present_Trento.pdf

Thursday 10 July

09:00 → 09:50 Towards modelling DNA in biological environments

When aiming for atomic detail, DNA has typically been characterized using short, relaxed fragments (up to 50 base pairs or bp). However, inside cells, DNA is a very long polymer that is supercoiled, subjected to bending and pulling forces, and surrounded by a crowded environment. In our lab, we have developed approaches and protocols for simulating DNA at the atomic level while accounting for these mechanical stresses in order to recreate more realistic conditions. In my presentation, I will show how DNA structure and dynamics respond to supercoiling when combined with pulling and DNA-bending proteins (1-4). Furthermore, I will provide atomistic insights into the non-specific interactions between some of the most abundant proteins in cells and DNA. The aim is to initiate a discussion regarding the potential acceleration of the study of these effects using AI.

1. ALB Pyne et al., Nat Commun, 12, 1053 (2021)
2. S Yoshua et al., Nuc Acids Res, 49, 8684-8698 (2021)
3. GD Watso et al., Comput Struct Biotech J, 20, 5264-5274 (2022)
4. M Burman and A Noy, Phys Rev Lett, 134, 038403 (2025)

Speaker: Agnes Noy (University of York)

09:50 → 10:20 TBD

Speaker: Dusan Racko (Polymer Institute, Slovak Academy of Sciences)

10:20 → 11:05 Coffee break

11:05 → 11:35 Everything everywhere all at once in enhanced sampling of rare events

Most natural processes, due to the large free energy barriers that separate the metastable states, are rare events for computer simulations, making their study challenging.
To alleviate this issue, a number of enhanced sampling methods have been proposed.
Methods such as Metadynamics or On-the-fly Probability Enhanced Sampling (OPES)[1,2] aim at removing the effect of barriers to promote transitions between states by filling the energetic basins. This way, at convergence, the sampling of the phase space can be improved and free energy estimates recovered.
However, such methods require the definition of collective variables that encode the relevant mode of the process, which are not always easy to obtain.
In addition, such simulations hardly sample the crucial transition state (TS) region, which provides precious insights into the workings of the reactive process.
Indeed, the TS is located on saddle points of the energetic landscape, making its sampling highly unfavourable also in the biased scenario.

Recently, we proposed an approach, based on machine learning the committor function through an iterative and self-consistent procedure, to extensively sample and analyze the TS.[3]
Such a method is based on a bias potential, functional of the committor, that helps stabilize the TS region, thus favoring its sampling.

We have then greatly improved this procedure by combining it with a metadynamics-like (OPES[1,2]) enhanced sampling approach, using a logarithmic function of the committor as a collective variable.[4]
The combined action of the two biases leads to a much improved sampling in which transition and metastable states are studied with the same thoroughness, transitions are efficiently observed, and accurate free energy estimates recovered.

After the first test systems, including small proteins and host-guest systems, the method, whose code is available open-source[2], is already being applied to speed up and increase the level of detail in challenging applications in biophysics, ranging from protein dynamics driven by force-fields to enzymatic reactions at QM/MM level[5].

[1] M.Invernizzi, M. Parrinello, JPCL, 7, 2731–2736, 2020
[2] E.Trizio et al., ArXiv preprint, 2024
[3] P.Kang, E.Trizio, M.Parrinello, Nat.Comp.Sci. 4, 451–460 2024
[4] E.Trizio, P:Kang, M.Parrinello, Nat.Comp.Sci. 5, --- 2025
[5] S.Das et al., ChemRxiv preprint, 2025

Speaker: Enrico Trizio (Italian Institute of Technology)

11:35 → 12:05 Cooperative Binding and the Control of DNA Replication Timing in Bacteria

Bacteria have evolved a sophisticated strategy to coordinate DNA replication with cell growth, enabling them to grow rapidly in nutrient-rich conditions even when the cell doubling time is shorter than the duration of a single round of replication. However, the precise mechanism by which the replication program is coupled to the cell cycle remains debated.

Most bacteria have a circular chromosome with a single origin of replication, where the DNA double helix is unwound and replication begins. Initiation is triggered by the protein DnaA, whose accumulation in an active, initiation-competent state is tightly regulated. Two major processes control this regulation: the titration at chromosomal binding sites, and the cycling between the active and inactive forms of DnaA.

DnaA binds to high-affinity sites distributed along the chromosome, effectively reducing the concentration of free, unbound DnaA. Although DnaA is continuously produced such that its total concentration remains approximately constant throughout the cell cycle, when replication is not occurring, the concentration of chromosomal binding sites decreases due to volume expansion. As these titration sites become saturated, the accumulated free DnaA increasingly binds to sites at the origin to trigger initiation. During replication, new binding sites are created as the DNA strands are copied, which in turn reduce the free DnaA concentration, making premature re-initiation unlikely.

However, this feedback alone is not sufficient to ensure robust control of initiation timing. An additional layer of regulation is provided by the ATP-dependent inactivation of DnaA following initiation.

While replication initiation is inherently stochastic, bacterial cells nonetheless achieve remarkably accurate timing and maintain synchrony of origin firing across multiple replication forks. We present a statistical model based on cooperative binding of DnaA at the origin to explain how bacterial cells achieve such precise replication timing despite the intrinsic stochasticity of the underlying molecular processes

Speaker: Alberto Stefano Sassi (Okinawa Institute for Science and Technology)

Sassi_Trento_July_2025.pdf

12:05 → 14:30 Lunch break

14:30 → 15:20 TOPAS-nBio – a Monte Carlo approach to mechanistic modeling at the cell-scale, connecting physics and biology

Purpose: TOPAS-nBio brings a track-structure Monte Carlo (MC) simulation framework to the research community to test hypotheses of radiation effects at the nanometer/cell scale. Here, I present the developments and progress made over the last decade of the TOPAS-nBio project.

Methods: TOPAS-nBio1 extends the TOPAS2 MC application to the nanoscopic scale, and is built on the Geant4 MC toolkit, an open-source MC framework for radiation transport. The TOPAS-nBio project links detailed MC track-structure simulations with geometrical representations of (sub-) cellular components, including initial chemical processes and likely biological outcomes via mechanistic models of DNA repair. The DNA is modeled as a chain of molecules with a resolution of single base (~0.34 nm).

Results: The latest TOPAS-nBio release v4.0 (https://github.com/topas-nbio) provides a simulation framework for nanometer scale radiobiology research. We have developed a variety of geometries, including multiple cell topologies and different representations of DNA geometries, including cell-line specific DNA representations using Hi-C data. Energy depositions can be scored to obtain induction of direct DNA damage after irradiation or indirect DNA damage following chemical reactions after radiolysis. Special emphasis was given to the chemistry framework, improving simulation speed using the independent reaction time method, expanding the simulated time to include long-term (seconds to hours) reactions by merging non-homogeneous and homogeneous chemistry stages, adding chemistry processes mimicking cell environments, including reactions with DNA constituents, and the capability to simulate ultra-high dose-rate irradiation. Two models of DNA repair kinetics have been linked to predict final biological effects and cell fate.

Conclusion: TOPAS-nBio offers cross-disciplinary simulations aiming to understand cell-level radiobiology, and has been applied internationally by multiple groups, providing insights into mechanisms of cell responses for different radiation modalities or when investigating new technologies (e.g., radiosensitization with nanoparticle or healthy tissue protection with ultra-high dose rate irradiations).

References:
1 B. Faddegon et al., Phis. Med., 72, 114-121 (2020).
2 J. Schuemann et al., Radiat. Res., 191(1), 125-138 (2019).

Speaker: Jan Schuemann (Massachusetts General Hospital & Harvard Medical School)

15:20 → 15:50 Influence of DNA supercoiling on the kinetics of DSBs rupturing process

Double-strand breaks (DSB) involve the covalent cut of the DNA backbone over both strands and are widely recognized as the most detrimental lesions associated with the effect of ionizing radiation on cells: Particularly, their misrepair can lead to severe biological consequences, including genomic instability, apoptosis, or carcinogenesis.
DSBs can arise from a variety of processes that range over broad (but intertwined) spatial and temporal scales, making their early-stage characterization challenging to conventional experimental techniques. On the other hand, in silico approaches have proven to achieve valuable insights in this context. Indeed, Monte Carlo track-structure simulations and microdosimetric models have successfully correlated the early effects of cell irradiation with macroscopic biological outcomes (i.e. the cell survival) by providing mean-field descriptions of radiation fields. However, these methods neglect the structural and mechanical dynamics of damaged DNA at the molecular level, which is largely overlooked despite its biological relevance.
In this presentation, we show how coarse-grained molecular dynamics simulations has been employed to characterize the DNA rupturing process in supercoiled DNA minicircles, lesioned by various DSB motifs: Particularly, we systematically explore how topological, structural, and mechanical features influence the rupture kinetics.
Our results reveal that mechanically-strained DNA conformations overall exhibit a higher rupturing probability than their topologically-relaxed counterparts - this effect being significantly enhanced under positive supercoiling regimes - despite highlighting a topological asymmetry in the mechanical response of the DNA to the diverse lesions. Furthermore, the influence of topological and structural features on the DNA rupturing dynamics seemingly decreases over time, as DSBs relieve the topological constraints of circular DNAs and drive the structural relaxation of the excess supercoiling stress.
In conclusion, our findings tally with previous radiobiological observations suggesting that compact DNA conformations reduce its radiosensitivity by minimizing the effective target volume. Moreover, we infer that DNA molecules exhibit minimal rupture enhancement at a biologically-relevant negative superhelical density (σ=−0.06), suggesting that such regime represents a favorable state against exogenous damage. Overall, these results support the idea that negative DNA supercoiling, beyond its roles in gene regulation and genomic organization, may also provide incidental structural protection against radiation-induced damage.

Speaker: Manuel Micheloni (University of Trento)

ManuelMicheloni.pdf

15:50 → 16:20 A fast GPU Simulations of Multicellular Response to Radiation Effects on Cell Populations

We a present novel multiscale simulation framework for describing and predicting radiation-induced biological damage considering both the slow biological and the fast chemical scales.
Building upon the Generalized Stochastic Microdosimetric Model (GSM2) [1] a fully probabilistic model for DNA damage formation and kinetic evolution, we develop a hybrid mesoscopic methodology that include cell proliferation and division besides the already existing description of DNA lesion formation and repair
The code, developed in Julia, is designed to adapt easily to various cell population geometries and arbitrary 3D cell distributions. It uses Monte Carlo sampling to simulate cell damage induced by radiation including proton, helium or carbon, beams and calculates the survival probability and the time of repaire or apoptosis, for each individual cell.
We bridged this code with CellSim3D [2], an open-source code developed in CudaC, to simulate cells evolution and division. The radiation Julia module establishes an initial state, which will then serve as input of a modified version of CellSim3D, integrating a reparation mechanisme.
We then build a simulation pipeline allowing use to study temporally fractionated radiotherapy. This simulation pipeline enables the study of temporally fractionated radiotherapy allowing to include any temporal and spatial fractionation schemes and radiation quality. We are now able to simulate radiation exposure and three days of cellular evolution in a millimeter-scale volume of densely packed cells in under 20 minutes.
[1] Generalized stochastic microdosimetric model: The main formulation, F. Cordoni, M. Missiaggia, A. Attili, S. M. Welford, E. Scifoni, and C. La Tessa, Phys. Rev. E 103, 012412, 2021.
[2] CellSim3D: GPU Accelerated Software for Simulations of Cellular Growth and Division in Three Dimensions Madhikar, P., Åström, J., Westerholm, J., & Karttunen, M. (2018). Computer Physics Communications, 232, 206–213

Speaker: Jules Morand

16:20 → 17:05 Coffee break

17:05 → 18:30 Poster session

Friday 11 July

09:00 → 09:50 Bridging Scales in Molecular Simulations: Data-Driven Characterization of Free Energy Landscapes

Understanding biomolecular behavior across multiple spatial and temporal scales remains a central challenge in molecular biophysics. Atomistic simulations offer fine-grained detail, but are computationally expensive, while coarse-grained models provide efficiency at the cost of resolution. Bridging these scales requires robust frameworks to translate between models and ensure consistent thermodynamic behavior across representations.

In this talk, I will present recent work on data-driven characterization of free energy landscapes, highlighting how machine learning and dimensionality reduction techniques can be used to identify relevant collective variables and thermodynamic basins from simulation data. I will then discuss how these insights can be transferred between atomistic and coarse-grained descriptions, enabling multiscale modeling of complex systems such as multy-body biomolecules.

Speaker: Oleksandra Kukharenko (Max Planck Institute for Polymer Research)

09:50 → 10:20 Enhanced Artificial Intelligence Molecular Mechanism Discovery of Reaction Paths in Complex Systems

Self-organization of molecules into ordered structures is crucial for both living and non- living matter. Artificial Intelligence Mechanistic Molecular Discovery[1] is an autonomous transition path sampling algorithm that uses deep learning to discover the underlying reaction mechanism of such molecular self-organizing phenomena. The algorithm uses the outcome of unbiased dynamical trajectories to construct, validate and update the quantitative mechanistic model. With the learned mechanistic model, the sampling of mechanistic trajectories or rare events can be enhanced, completing the cycle. Here, we enhance the method by including all potentially possible configurations in the reweighted path ensemble. We illustrate the novel methodology on simple potentials and a more complex molecular system.

[1] Jung et al., Nat. Comput. Sci. 3, 334–345 (2023).

Speaker: Rik Breebaart (University of Amsterdam)

10:20 → 11:05 Coffee break

11:05 → 11:35 Percolation in Excitable Media: Insights into Signal Propagation in Heart-like Systems

Understanding how electrical signals propagate through biological tissues is central to the study of physiological function and pathological breakdown. In this work, we explore a model system inspired by cardiac tissue, using tools from percolation theory and statistical physics to investigate the emergence and failure of global signal propagation. Our model captures the interplay between structural connectivity and dynamic excitability, revealing critical thresholds beyond which the system loses its capacity to sustain coordinated activity.

We simulate a two-dimensional excitable medium, mimicking heart-like networks, where local inhomogeneities and quenched disorder lead to complex propagation patterns. Through percolation analysis, we identify how microscopic disruptions—whether due to blocked pathways or inactive nodes—can lead to macroscopic breakdowns in signal transmission. The results offer insight into phenomena such as conduction block and fibrillation, which are of clinical relevance.

While rooted in theoretical physics, our study bridges concepts between statistical mechanics and biological dynamics, emphasizing how simple models can offer deep understanding of complex physiological systems. This framework, although classical, lays the groundwork for further integration with data-driven or AI-enhanced models in biological tissue analysis.

Speaker: Md Aquib Molla (Vidyasagar College)

Md_Aquib_Molla_Percolation_Talk.zip

11:35 → 12:05 Analysis of protein folding through the combination of transition path theory and informative low-resolution representations

The biologically active conformation of a protein is maintained by the interactions among those residues involved in native contacts, but in the course of the folding process amino acids can be involved in transient native as well as non-native interactions, which drive the molecule towards its folded state. Reconstructing how these interactions evolve is key to understanding the folding process; however, no technique is currently available to do so in a general and unsupervised manner.
We here present a method that highlights the most functionally relevant residues of a protein at each stage of its folding pathway. To do so, we jointly leverage two distinct techniques: transition path theory (TPT), to decompose the folding pathway in a sequence of steps according to the committor function; and the mapping entropy optimization workflow (MEOW), which highlights, at each step, the subset of residues that play the most relevant structural, energetic, and functional role.
We validate this method by applying it to a benchmark yet nontrivial case, namely miniprotein chignolin, showing that the combination of TPT and MEOW provides novel information on the molecule’s folding pathway that is nonetheless coherent with well-established results. This approach, which is of general applicability, complements existing analysis methods and paves the way to an increasingly detailed comprehension of protein folding.

Speaker: Alessia Guadagnin Pattaro (University of Trento)

12:05 → 12:20 Conclusive remarks