News: Research
News Type: All | Research | Education & Outreach | People | Announcements
This work introduces a novel learning paradigm—Temporal Contrastive Learning (TCL)—that enables physical systems to learn autonomously without centralized control or explicit memory . In a recent work published in Nature Communications by Martin Falk, Murugan lab at the Physics Frontier Center for Living Systems, and Adam Strupp, a framework was proposed where integral feedback dynamics and alternating temporal protocols (free vs. clamped phases) generate an implicit memory necessary for contrastive learning. Unlike backpropagation, which requires global coordination, TCL operates through local, time-dependent updates, making it suitable for non-equilibrium physical substrates such as chemical, mechanical, or neuromorphic systems. The authors quantify the energy dissipation associated with learning and derive a Landauer-like bound on its thermodynamic cost. This work advances the physics of learning by showing how non-equilibrium dynamics and feedback control can endow physical systems with computational capabilities, bridging statistical mechanics, machine learning, and information theory.
Microbial ecosystems are governed by complex, often hidden, interactions driven by competition for shared resources. In this work, Xiaowen Chen and Kyle Crocker from Seppe Kuehn’s group collaborated with Thierry Mora and Aleksandra Walczak to introduce a physics-based framework that applies spectral analysis of time-series data to infer the structure of resource competition in microbial communities. By leveraging tools such as cross-power spectral density (CPSD) and coherence, the authors show that time-delayed correlations—rather than static ones—more accurately capture competitive interactions. Using both synthetic consumer-resource models and real-world plankton data, they demonstrate that these spectral signatures reveal metabolic guilds and interaction networks aligned with genomic similarity. This study advances biological physics by applying principles from non-equilibrium statistical mechanics and signal processing to decode ecological structure from dynamic data, offering a powerful new lens for understanding microbial diversity and function.
Biological systems adapt to their environments through feedback between mechanical forces and biochemical activity. Postdoctoral fellow Deb Banerjee from Suri Vaikuntanathan’s group led a project in collaboration with Martin Falk (Murugan), Margaret Gardel, Aleks Walczak and Thierry Mora to show that cytoskeletal networks—composed of force-sensitive proteins and active molecular motors—can implement a biophysically grounded learning mechanism to realize adaptive mechanics. Using a coarse-grained model of the actomyosin cytoskeleton, they show that such networks can undergo contrastive learning in response to environmental perturbations, adjusting their internal structure to enable learning of desired mechanical responses. The study provides a minimal, biologically plausible model of learning in a material framework, advancing understanding of non-equilibrium mechanics, information encoding, and adaptive behavior in living systems. It opens new directions for understanding how physical systems can perform computation without a central processor.
Natural environments are rarely flat. Walking animals must navigate substrates that are rough and variable across multiple scales. We do not understand the principles of adaptability in biomechanics that can allow an organism to deal with such variability on a range of length scales. We showed how moth flies (Clogmia albipunctata) adjust stepping kinematics and coordination patterns to deal with substrates of varying levels of roughness. Above is an experimental image of a moth fly walking across sand paper of a certain grit. The gait changes as grit is modified. Colored dots are features used to track gait. This work is published by Brandt et. al., Annals of the New York Academy of Sciences 2024 from Nirody’s lab at the Physics Frontier Center for Living Systems.
03/20/2024
Pattern formation by turbulence
Pattern formation entails a process of wavelength selection, which can usually be traced to the linear instability of a homogeneous state. By contrast, here we propose a fully nonlinear mechanism triggered by the non-dissipative arrest of turbulent cascades. We show that the tunable wavelength of these cascade-induced patterns can be set by a non-dissipative transport coefficient called odd viscosity. Such cascade-induced patterns naturally arise across scales, from bio-active chiral fluids to stellar plasma or in the coagulation of droplets in which mass rather than energy cascades. This work was published by de Wit et. al. Nature 2024 in the Vitelli group at the Center for Living Systems.
Cellular form and function emerge from complex mechanochemical systems within the cells. Currently, no systematic strategy exists to infer large-scale physical properties of a cell from its molecular components. This is an obstacle to understanding processes such as cell adhesion and migration. Here, we develop a data-driven modeling pipeline to learn the mechanical behavior of adherent cells. We developed a data-driven modeling pipeline to learn the mechanical behavior of adherent cells. We first train neural networks to predict cellular forces from images of cytoskeletal proteins. We then develop two approaches—one constrained by physics and the other agnostic—to construct data-driven continuum models of cellular forces. Both reveal how cellular forces are encoded by two distinct length scales. Beyond adherent cell mechanics, our work serves as a case study for integrating neural networks into predictive models for cell biology. This work, published in Schmitt et al., Cell 2024 is a collaboration between Vitelli and Gardel.
Spiking is a general phenomenon that is crucial in the firing of neurons, beating of hearts, and spread of diseases. In homogeneous media, spiking arises from a local competition between amplifying and suppressing forces. But most real-world systems are far from homogeneous. We demonstrated that inhomogeneities such as interfaces and boundaries (that spatially segregate these two forces) can promote spiking, even if the system does not spike when these forces are evenly mixed. Our findings apply to chemical reactions, predator–prey dynamics, and recent electrophysiology experiments, in which localized action potentials were observed at the interface of distinct, nonspiking bioelectric tissues. This work was published by Scheiber et. al. PNAS 2024.
Life on earth relies on sunlight for energy, but this energy can only be exploited through the collective recycling of matter by communities of microbes, plants, and animals. Yet we lack a framework for understanding how ecosystems can organize themselves to collectively capture the sun’s energy by running cycles of matter subject to thermodynamic constraints. We advance a conceptual model to study the collective properties of nutrient-cycling ecosystems. Surprisingly, even though species “greedily” extract energy from the environment, sufficiently diverse communities of species almost always manage to sustain themselves by extracting enough energy. The plots above show distributions of the total energy extracted by ecosystems as a function of the number of species. As ecosystems become more species-rich, ecosystems extract greater average energy with greater convergence. This result from the Murugan lab at the CLS is described in Goyal et al, PNAS (2023).