A coarse-grained molecular dynamics framework used to simulate plasmid DNA analyzed via atomic force microscopy (AFM). The resulting images are used to train a U-Net for DNA chain and crossing segmentation and classification. (Image by P. Dutta.)ASAP (AFM Simulation and Analysis Pipeline)
Prakhar Dutta, Jiacheng Huang, Nazli Demirpehlivan, Thomas Catley, Sylvia Whittle, Carlo Manzo, Rahul Nagshi, Rachel Owen, Giovanni Volpe Date: 11th March 2026 Time: 18:00 – 20:00 Place: Aula Medica, Karolinska Institute, Solna
Conference Protein Folding in Real Time, 11-13 March 2026, Stockholm, Sweden
Abstract: Atomic force microscopy (AFM) resolves biological structure and mechanics at high resolution, but produces vast, heterogeneous datasets that are often noisy and very time-consuming to analyse. Although deep learning could automate quality control, segmentation and feature extraction, adoption is limited by scarce ground-truth training data and high technical barriers for experimentalists. Here we present ASAP, an open-source tutorial and pipeline implemented in DeepTrack to provide a reproducible foundation for AI-enabled AFM. At the protein folding conference, a dual-pathway simulation for DNA, offering both molecular dynamics and rapid, non-MD geometries to generate perfect ground truth for segmentation training was presented. By consolidating simulation and learning into a single modular ecosystem, this work enables users to build upon our pipeline to optimize AFM workflows for more efficient data acquisition and robust processing.
Reconstructed field of LCD2–CTPR4-Func1 condensates with LC3 at the sample plane (shown here as the imaginary component of the complex field). The condensates increase in size through Ostwald ripening and recruitment of LC3. (Image by X. Zhang.)Quantitative Characterization of Biomolecular Condensates Using Off-Axis Holographic Microscopy
Xinwen Zhang, Nora Haanaes, Berenice García Rodríguez, Giovanni Volpe, Janet Kumita and Daniel Midtvedt Date: 11 March 2026 Time: 18:00-20:00 Place: Aula Medica, Karolinska Institute, Solna
Conference Protein Folding in Real Time, 11-13 March 2026, Stockholm, Sweden
Biomolecular condensates formed through liquid–liquid phase separation (LLPS) play important roles in cellular organization, yet quantitative and label-free characterization of their physical properties remains challenging. In this work, we apply off-axis holographic microscopy to study a synthetic biomolecular condensate platform based on the LCD2-CTPR protein system. These proteins, composed of modular consensus-designed tetratricopeptide repeat (CTPR) domains fused to intrinsically disordered regions, undergo phase separation under varying salt concentrations. By incorporating short binding motifs such as ATG13 or Func1, the condensates can specifically recruit the autophagy-related protein LC3. Using label-free quantitative phase measurements, we analyze changes in condensate optical radius and refractive index during LC3 recruitment and over time. Our results show measurable variations in condensate size and optical properties, highlighting the sensitivity of these systems to compositional changes. This work demonstrates the applicability of holographic microscopy for quantitative characterization of synthetic biomolecular condensates and provides a framework for studying protein phase separation in a non-invasive manner.
Pointwise absolute error plots for the heat equation tested across five architectures. The plots show the mean absolute error achieved by each architecture on a periodic-mode initial condition. Errors are shown on a logarithmic scale. Blue colors indicate smaller errors. (Image by E. A. Duta Costache.)The optimization autopsy of PINNs
Eduard Duta Costache, Benjamin Girault Date: 11 March 2026 Time: 18:00-20:00 Place: Aula Medica, Karolinska Institute, Solna
Conference Protein Folding in Real Time, 11-13 March 2026, Stockholm, Sweden
Physics-Informed Neural Networks (PINNs) have emerged as a promising method for solving Partial Differential Equations (PDEs) by combining data-driven learning with physical laws. However, the spectral bias and optimization challenges limit their efficacy. This work investigates these issues and whether the advantages of classical spectral methods translate to the non-convex neural network optimization landscape. We show that gradient imbalance greatly affect learning and we study the Hessian conditioning under different settings. Our results indicate that spectral priors stabilize training, reduce error, and improve parameter efficiency. We also identify that learnable-basis models act as implicit regularizers under sparse sampling.
Recent advances in nonequilibrium physics allow extracting thermodynamic quantities, such as entropy production, directly from dynamical information in microscopic movies. (Figure by S. Manikandan, adapted from Manikandan et al., Phys. Rev. Research 6, 023310 (2024).)Probing the Non-equilibrium Dynamics of Living Matter
Sreekanth K Manikandan Date: 11 March 2026 Time: 18.00-20.00 Place: Aula Medica, Stockholm Sweden
Conference Protein Folding in Real Time, 11-13 March 2026, Stockholm, Sweden
Identifying whether a process is in equilibrium, quantifying its distance from equilibrium, and constructing optimal reduced descriptions of non-equilibrium dynamics remain central challenges in the study of living matter. Here, we discuss how data-driven approaches grounded in stochastic thermodynamics enable these features to be inferred directly from experimental data. In particular, we show how entropy production can be localized in space and time, and how maximally dissipative coordinates emerge as effective low-dimensional descriptions of non-equilibrium processes. We highlight applications to experimental biophysical systems and discuss key challenges and limitations.
Schematic representation of the architecture of a quantum circuit for application in variational problems. (Image by A. Schiano di Colella.)Quantum computing for variational problems
Andrea Schiano di Colella, Antonio Ciarlo, Mats Granath, Giovanni Volpe Date: 11 March 2026 Time: 18:00 – 20:00 Place: Aula Medica, Karolinska Institutet, Stockholm, Sweden Conference: Protein Folding in Real Time
Quantum computing is a field of study that aims to exploit quantum mechanical effects for the purposes of computation. Due to the intrinsic capacity of qubits of efficiently represent an exponentially large configuration space, quantum computation has been identified as a promising candidate for complex physical chemistry simulations, including investigating the dynamics of protein folding. This work illustrates the use of quantum computing for variational problems, and the use of alternative training methods such as genetic algorithms to avoid the “barren plateau” phenomenon, which prevents the training of general quantum circuits by means of the usual gradient descent.
3D model of the integrated setup combining optical tweezers, light-sheet microscopy, and microfluidics to manipulate the gut microbiome in vivo in zebrafish. (Image by N. C. Palmero Cruz.)Optical Manipulation of Gut Microbiome and Neural Responses in Zebrafish
Norma Caridad Palmero Cruz Date: 11 March 2026 Time: 18.00-20.00 Place: Aula Medica, Stockholm Sweden
Conference Protein Folding in Real Time, 11-13 March 2026, Stockholm, Sweden
The gut-brain axis is a complex, bidirectional network linking the microbiome to the central nervous system, significantly affecting physiological processes and neurological health, including conditions like autism and depression. Due to the genetic similarities between zebrafish and humans, the zebrafish serves as a valuable model for investigating the bidirectional relationship between the gut and brain, offering insights into how it compares with human behaviors. Research on the connection between gut and brain development typically involves using germ-free lab animals, where the microbiome is eliminated, and comparing them to those with restored microbiomes. However, this method does not capture the complexity of microbiome-nervous system communication due to its all-or-nothing approach. This work presents a setup that combines microfluidic techniques, optical tweezers, and light sheet microscopy to precisely manipulate the microbiome in larval zebrafish in situ and in vivo. This approach offers deeper insights into gut-brain connectivity and its impact on neurological health.
Heat-induced aggregates in Saccharomyces cerevisiae on Slimfield SMLM. Hsp104-mGFP binds to misfolded regions enabling aggregate visualisation for tracking. (Image by L. Viaene.)A single-molecule approach to study the spatial protein quality control system
Linde Viaene Date: 11 March 2026 Time: 18.00-20.00 Place: Aula Medica, Stockholm Sweden
Conference Protein Folding in Real Time, 11-13 March 2026, Stockholm, Sweden
Cell populations are inherently diverse, and averaging measurements across them can mask subtle or rare cellular behaviours. In this work, we use live Slimfield single-molecule microscopy to study the role of Hsp104 in clearing misfolded and aggregated proteins after stress. By analysing endogenously tagged Hsp104, we quantify molecular diffusion and stoichiometry before and after heat stress. Our results show a transition from faster, more mobile molecules to larger, more static assemblies following stress, consistent with Hsp104 functionally engaging with protein aggregates. These measurements provide molecular-level insight into how cells respond to proteotoxic stress.
Identifying whether a process is in equilibrium, quantifying how far it lies from equilibrium, and determining optimal reduced descriptions of non-equilibrium processes remain challenging open problems. Here, we discuss how novel data-driven techniques grounded in stochastic thermodynamics can be used to efficiently learn these features directly from experimental data. In particular, we show how entropy production can be localized in space and time, and how maximally dissipative coordinates can be consistently inferred as effective low-dimensional descriptions of non-equilibrium processes. We further discuss applications to experimental biophysical systems and outline key challenges and limitations.
Schematic illustration of the light-momentum detection principle underlying SmartTrap. The momentum change of the trapping laser, induced by its interaction with the trapped particle, is measured to directly quantify optical forces with high precision, enabling real-time feedback and autonomous control in non-equilibrium experiments. (Figure by A. Ciarlo.)SmartTrap: Autonomous Optical Tweezers for Statistical Physics of Non-Equilibrium Systems
Antonio Ciarlo Date: 26th February 2026 Time: 13.30 Place: NORDITA, Stockholm, Sweden The 15th Nordic Workshop on Statistical Physics: Biological, Complex and Non-equilibrium Systems
Optical tweezers are a key tool in non-equilibrium statistical physics, allowing direct measurements of forces, work, and fluctuations in single-molecule and soft matter systems. However, manual operation limits throughput and the systematic study of rare events.
In this talk, Antonio Ciarlo will present SmartTrap, a fully autonomous optical tweezers platform integrating deep learning–based 3D tracking, adaptive feedback control, and automated microfluidics. The system operates without human intervention, executing complete force spectroscopy protocols.
Demonstrated with high-throughput DNA pulling experiments on λ-DNA, SmartTrap enables precise measurements of force–extension curves and folding kinetics. The platform also opens new possibilities for studies of colloids, single cells, and quantitative tests of non-equilibrium statistical physics.
Active Matter: Model Systems and Experimental Tests
Agnese Callegari, Antonio Ciarlo, Sreekanth Manikandan Dates and times:
23 Feb 14:00-15:00 (Agnese)
24 Feb 11:30-12:30 (Antonio)
24 Feb 14:00-15:00 (Sreekanth) Place: PJ Winter school on Geometry of nonequilibrium critical phenomena
Active matter is a broad class of systems that operate intrinsically out of equilibrium. It spans multiple length scales—from macroscopic to micro- and nanoscopic—and includes both biological and artificial realizations, often displaying rich and emerging collective behaviors. The study of active matter aims to explain and interpret these phenomena using concepts and tools from physics. As such, understanding active and non-equilibrium systems requires a combination of theoretical, computational, and experimental approaches.
In the first part of the lecture, we introduce the concept of an active particle and demonstrate how it can be embodied in a macroscopic, self-propelled toy robot (a Hexbug). Despite their simplicity, such systems reproduce characteristic—and sometimes counterintuitive—features of microscopic active matter. These experiments have a strong pedagogical value and are designed to help bridge a gap in traditional physics curricula at the primary and secondary education levels.
The second part of the lecture focuses on active matter and non-equilibrium phenomena at the microscopic scale, where advanced experimental tools are essential. Optical tweezers provide precise control over microscopic systems and access to key physical observables. We introduce their operating principles and illustrate how they can be used to construct a minimal, well-controlled experimental model for studying non-equilibrium dynamics at the single-particle level.
In the final part of the lecture, we turn to the theoretical and computational tools required to analyze active matter systems. We discuss how non-equilibrium dynamics can be quantitatively characterized directly from experimental data in a model-independent framework. This naturally leads to an introduction to machine-learning–based inference techniques, which extract dynamical and thermodynamic information from data without relying on a priori assumptions about the underlying physical model.
References:
[1] A. Barona Balda, A. Argun, A. Callegari, G. Volpe. Playing with Active Matter, Am. J. Phys. 92, 847–858 (2024). https://doi.org/10.1119/5.0125111
[2] Martins, T.T., Malavazi, A.H.A., Kamizaki, L.P. et al. Fluctuation theorems with optical tweezers: theory and practice. Eur. Phys. J. Plus 141, 71 (2026). https://doi.org/10.1140/epjp/s13360-025-07181-4
[3] Manikandan, Sreekanth K. and Ghosh, T. and Mandal, T. and Biswas, A. and Sinha, B. and Mitra, D. Estimate of entropy production rate can spatiotemporally resolve the active nature of cell flickering. Phys. Rev. Res. 6, 023310 (2024). https://doi.org/10.1103/PhysRevResearch.6.023310
Photos
Antonio, presenting. (Photo by M. Orsino)Sreekanth, presenting. (Photo by A. Ciarlo)
