Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session T10: Physics of Learning: Learning and Adaptation in Living Systems |
Hide Abstracts |
Sponsoring Units: DBIO Chair: Pradeep Natarajan, Massachusetts Institute of Technology Room: Room 202 |
Thursday, March 9, 2023 11:30AM - 11:42AM |
T10.00001: Information processing by heterogeneous cell populations Purushottam Dixit Cells’ responses to dynamic changes in the environment are corrupted by inherent stochasticity in intracellular signaling networks. The mutual information (or its maximum, the channel capacity) quantifies the fidelity of cellular signaling. However, current approaches to estimate channel capacity average over the extensive heterogeneity in cell state variables that typically characterizes cell populations. To explicitly account for cell-to-cell differences in quantifying signaling network fidelity, we develop a novel information theoretic framework, cell-state conditioned mutual information (CeeMI). CeeMI models individual cells in a population as unique information channels and estimates the average mutual information between the input and the output as an average over cell state variables. We estimate CeeMI for two signaling pathways and show that it is significantly larger than traditional estimates. Using the IGFR/FoxO pathway, we verify that that individual cells can differentiate between multiple levels of environmental stimuli. Importantly, our approach allows us to identify intracellular biochemical parameters that make some cells good and others bad and sensing their environment. We believe that our new framework will significantly improve our understanding of how cells detect changes in their environment. |
Thursday, March 9, 2023 11:42AM - 11:54AM |
T10.00002: Optimal cellular adaptation in fluctuating metabolic microenvironments Jason T George Cellular metabolic adaptation defines how living systems navigate highly variable nutritional environments, from single cells in spatiotemporally varying environments to complex multicellular organisms anticipating the next nutrient-rich period. Given the relevance of metabolic adaptation to many chronic diseases, including diabetes and cancer, the precise nature by which adaptation occurs is of great interest and not fully understood. Here, we develop a stochastic model to characterize the decision-making employed by systems that optimally navigate fluctuating nutritional environments. In accounting for experimentally observed cellular memory, we show that adaptive populations capable of sensing their environments and estimating the nutrient state more effectively navigate fluctuating metabolic environments when compared to their passive counterparts. We demonstrate the benefit of larger metabolic memory on long-term growth rates for stationary environments, and we find cells capable of adjusting their memory can more efficiently grow in a changing metabolic landscape. Surprisingly, when comparing the growth rate of adaptive cells in constant environments to those in random environments, our model predicts that the later population does consistently worse, in agreement with recent empirical observations in bacterial systems. Our modeling framework is of general utility for studying phenotypic responses to signaling input schemes requiring systems to strike a tradeoff across multiple phenotypes and fluctuating environmental states. Such modeling can be applied to predict the response of adaptive systems to environmental alteration and therapeutically relevant interventions. |
Thursday, March 9, 2023 11:54AM - 12:06PM |
T10.00003: One nose but two nostrils: interaction between two hemispheres aligns bilateral responses Bo Liu Odors are detected by olfactory receptor neurons in the nose, which project to the ipsilateral olfactory bulb (OB). Each olfactory cortex on one hemisphere receives olfactory information from ipsilateral OB, plus contralateral information from the other hemisphere cortex. Since the projection from OB to cortex is largely random, odor representations in two cortices are presumably independent. Surprisingly, experiments in mice showed that the two representations are aligned, suggesting the inter-hemispheric projections must be structured. But how such structure emerges remains an enigma. |
Thursday, March 9, 2023 12:06PM - 12:18PM |
T10.00004: Infomation and Evolution Sarah Marzen, Jonathon Soriano Ongoing research on mutation variability seeks to explain a quantifiable aspect of the evolutionary process. In this talk, we offer an explanation that asexual populations' mutation rates are modified by mutator alleles to maximize information about their population. We consider the Wright-Fisher evolutionary model as the environment-population channel and compute its channel capacity to be 2 bits per allele. In addition, we find a lack of support for our hypothesis through a comparison between the expected mutation rate distribution and Red Yeast mutation rate distribution. We propose a study to experimentally test that asexual populations do not maximize information about their environment. |
Thursday, March 9, 2023 12:18PM - 12:30PM |
T10.00005: Do E. coli care about single molecules? Henry H Mattingly, Keita Kamino, Thierry Emonet, Benjamin B Machta Berg and Purcell derived a fundamental limit on how accurately concentration can be sensed from the stochastic arrival of ligand molecules to a cell’s surface by diffusion. However, it has remained unclear to what extent molecule counting noise is a meaningful limitation on sensing accuracy and downstream functions in cells. Answering this question has been challenging, even for E. coli chemotaxis, one of the paper’s original motivations. This is in part because E. coli chemotaxis depends on information the cell “gathers” about the fluctuating, instantaneous time derivative of concentration, not the absolute concentration, as we recently demonstrated. Here, we introduce an information rate that quantifies how much behaviorally-relevant information (in bits) is encoded per unit time by the receptor associated kinases. We show that the stochastic arrival of single particles limits this rate, which in turn limits E. coli’s ability to climb shallow gradients. |
Thursday, March 9, 2023 12:30PM - 12:42PM |
T10.00006: Alternation emerges as a multi-modal strategy for turbulent odor navigation Nicola Rigolli, Gautam Reddy, Massimo Vergassola, Agnese Seminara While looking for the source of a scent, rodents and dogs exhibit a complex alternating behavior: they spend most of the time moving while their noses explore the ground, but sometimes they pause to sniff in the air, rearing on their hind legs or raising their heads. This strategy has been observed in rodents exposed to airflow, suggesting that alternation has a primary role during plume-tracking. To understand where and when it is more convenient to sniff in the air, we combine fully resolved simulations of turbulent odor transport and Bellman optimization methods for decision-making under partial observability. We show that an agent trained to minimize search time in a realistic odor plume exhibits extensive alternation coupled with the characteristic cast-and-surge behavior observed in insects. Alternation is related to casting and occurs more frequently far downwind of the source, where the likelihood of detecting airborne cues is higher relative to ground cues. Casting and alternation emerge as complementary tools for efficient exploration with sparse odor cues. Our Partially Observable Markov Decision Process algorithm trained over simplified odor statistics performs robustly in different conditions, suggesting that evolution may have allowed mammals to develop tools to navigate optimally in chaotic environments. Finally, an analytical model based on marginal value theory captures the interplay between casting, surging, and alternation. |
Thursday, March 9, 2023 12:42PM - 12:54PM |
T10.00007: Allostery for free? Scenarios for the evolutionary origin of allostery in proteins. Eric Rouviere, Olivier Rivoire, Rama Ranganathan Protein allostery underpins many molecular processes in the cell and long standing efforts have sought to understand the physical mechanisms of allostery. The evolutionary origin of allostery on the other hand remains difficult to explain. For instance, latent allostery, or allosteric behavior not part of the protein's native function, is commonly observed yet can not be explained by direct selection alone. What is the origin of this latent allostery? Here use simplified physical models of proteins to show that allostery can emerge, not from direct selection, but indirectly, from selection on functions of the active site, such as binding specificity. We find that such selections lead to a spatially extended region of mutationally sensitivity where mutations act allosterically to modulate the function of the active site from a distance. Models with such architecture are "primed" with latent allostery and can act as evolutionary starting points for allosteric regulation. |
Thursday, March 9, 2023 12:54PM - 1:06PM |
T10.00008: Optimization and variability can coexist Lee Susman From molecule counting in bacterial chemotaxis to photon counting in human vision, many biological systems function near the physical limits to their performance. Some see this optimality as a natural consequence of evolution, while others emphasize that evolution is not generally guaranteed to find even local optima. More concretely, microscopic parameters of many biological systems -- be it protein copy numbers or spatial locations of retinal photoreceptors, patterns and strengths of synaptic connections, or sensing thresholds in gene regulatory networks -- are highly variable, and this variability seems like prima facie evidence against optimization. Here we show that this intuition is misleading: in problems ranging from transcriptional regulation to neural computation, variability in parameters can coexist with near optimal performance. Abstracting away domain-specific details, we show that this is possible because the parameter spaces of real biological systems are high dimensional. We discuss the implications of these observations for a range of biological and computational examples. |
Thursday, March 9, 2023 1:06PM - 1:18PM |
T10.00009: Training Networks with Internal Prestress Ayanna Matthews, Sidney R Nagel, Margaret Gardel We use disordered elastic networks to model amorphous materials. These models often assume that the networks have zero prestress on the bonds so that each bond is at its unstretched length. However many systems, such as biopolymers, have significant internal stresses even in mechanical equilibrium. We are interested in analyzing how this prestress affects the ability of a network to be trained. We found that prestress hinders training because of two causes: (1) prestress hides the local stress information that is typically required for training, (2) removing prestressed bonds during training alters the force balance at each site, thus leading to a change in geometry when a bond is pruned. We will show that the first issue is easily handled by a more general training rule which extracts the necessary stress information for training. However, this protocol is limited to highly coordinated networks and those with low amounts of prestress. The second issue is more difficult to circumvent since it is difficult to evaluate how the alteration of a bond changes the distribution of internal forces. Beyond a limit in the magnitude of the prestress, altering the network geometry overwhelms our ability to train effectively. We analyze quantitatively how much prestress can be in a network before the ability to train is no longer viable. |
Thursday, March 9, 2023 1:18PM - 1:30PM |
T10.00010: Physical Limits on Galvanotaxis Ifunanya Nwogbaga, Brian A Camley, A Hyun Kim Eukaryotic cells of many types can polarize and migrate in response to electric fields via “galvano- |
Thursday, March 9, 2023 1:30PM - 1:42PM |
T10.00011: Bayesian Origins of Growth, Cooperation, and Inequality in populations of learning agents Jordan Kemp, Luis Bettencourt "The statistical dynamics of population growth, wealth, and inequality are related through their common description as multiplicative stochastic processes. In these systems, variations in growth rates (and fitness) determine selection, relative wealth, and incentives for cooperation. Despite its relevance for biological and social sciences, we still lack a general theory that explains the dynamics of growth rates in terms of agent adaptation to their environment, heterogeneities of traits, and other consequential behaviors. In this talk, we derive a general population dynamics of learning agents leveraging knowledge of their environments to grow and reinvest their resources. Growth rates emerge in the long-time limit as the mutual information between agents signals and states of the environment. We show that, with knowledge of past and present conditions only, sequential Bayesian inference is optimal for maximizing growth, thus formally associating fitness with information. We discuss how this framework can address problems of inequality in heterogeneous populations and lay the foundations for a unified general quantitative theory of social and biological phenomena such as the dynamics of cooperation via Hamilton's rule, and the effects of education on life history choices." |
Thursday, March 9, 2023 1:42PM - 1:54PM |
T10.00012: Information processing in the adaptive immune response Obinna A Ukogu, Armita Nourmohammad The adaptive immune system surveils a large distribution of antigens and implements a range of complex multi-cellular outcomes in response. Each antigen is characterized by features encoded in its physical structure and the dynamics of its source pathogen. Although systems immunology has catalogued the molecular interactions mediating the immune response, we lack an understanding of how the response is calibrated to antigen features. Furthermore, as the response is constrained by molecular noise and phenotypic variation in responding cells, it is unclear how the system delivers an appropriate and effective response. Here, using computational and analytic methods, we investigate how the network of agents – antigens, cytokines, naïve, effector, memory and regulatory cells – processes information about antigen features. We compare the information capacity of many network structures and identify biologically plausible networks that maximize information transmission in the adaptive immune response. A comprehensive understanding of these networks may be critical for the safe and robust application of perturbative immune therapies in cancer and other diseases. |
Thursday, March 9, 2023 1:54PM - 2:06PM |
T10.00013: Increased dynamic range as a driver of ASD neuronal and behavioral differences Yuval Hart Successful social interactions rely on combining robust perception, quick updating of prior beliefs, and learning and generalizing from feedback. Individuals with Autistic Spectrum Disorder (ASD) exhibit difficulties in their social interactions compared with the neurotypical population (NT). Studying ASD enables decomposing social interactions to their basic cognitive processes to infer fundamental computational principles of social interactions. ASD compared to NT show heightened discriminability between stimuli, higher variance in neuronal activity, slower updating, deficits in learning and generalization, and reduced encoding capacity. Here, we suggest that these myriad of differences stem from a computational principle that relies on the dynamic range of the neuronal population response. The dynamic range of a sensing system is the range of signal values for which the system is responsive. We show that an increased dynamic range in the neuronal population response accounts for the neuronal and behavioral differences seen in ASD, across all outlined tasks and conditions. We further specify a plausible biological mechanism for the increase in the dynamic range, namely increased heterogeneity in the half-activation point of individual neurons in ASD. These findings suggest that the dynamic range of the neuronal population serves as a key factor to support the cognitive processes underlying social interactions as well as a principled way to tune these in artificial agents. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700