Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session H66: Inference, Information, and Learning in Biophysics II |
Hide Abstracts |
Sponsoring Units: DBIO GSNP Chair: Benjamin Machta Room: BCEC 261 |
Tuesday, March 5, 2019 2:30PM - 2:42PM |
H66.00001: Reconstructing biological dictionaries: How does neural code translate into behavior? Damian Hernandez, Samuel Sober, Ilya Nemenman The problem of deciphering how low-level patterns (action potentials in the brain, amino acids in a protein, etc.) drive high-level biological features (sensorimotor behavior, enzymatic function) represents the central challenge of modern biology. There are no general methods for reconstructing such dictionaries from small data sets, and for building vocabularies of statistically significant, independent words in them. Here we derive a method for solving this class of problems, which we call the unsupervised Bayesian Ising Approximation (uBIA) and demonstrate its utility in deciphering the motor code in the pre-motor neurons of a songbird. From small data sets, we detect codewords that predict behavior. These words contains arbitrary number of precisely timed spikes, confirming that the motor code in this system is build from precisely timed multi-spike patterns. We show that distinct classes of such words are used preferentially in codes responsible for motor exploration versus exploitation, opening a window on how motor behaviors are controlled by the brain. |
Tuesday, March 5, 2019 2:42PM - 2:54PM |
H66.00002: Learning to Crawl Shruti Mishra, Willem Van Rees, Lakshminarayanan Mahadevan We combine a mechanical model for a segmented, soft-bodied crawler (Paoletti and Mahadevan, 2014) with a reinforcement learning algorithm for choosing the neuronal excitations. The crawler chooses the neuronal excitations based on a minimal description of the state of its own body, with the goal of moving forward in 1-D. The gait achieved by learning neuronal excitations in this manner depends on the mechanical properties of the crawler. For a regime of properties, the crawler achieves forward locomotion by means of a peristaltic wave, qualitatively similar to what is observed in D. melanogaster larvae. This provides a mechanism for how organisms may learn to achieve locomotion in the absence of a central pattern generator, or recover from injury. This work also suggests a way to explore actuation patterns for soft-robots in cases where the optimal actuation patterns may not be intuitive and provides a means for online learning while exploring an unfamiliar terrain. |
Tuesday, March 5, 2019 2:54PM - 3:06PM |
H66.00003: Non-Gaussian Bayesian theory of sensorimotor learning with multiple timescales Baohua Zhou, David Hofmann, Samuel Sober, Ilya Nemenman Various experimental studies have documented that sensorimotor learning phenomena occur on multiple timescales. Many theoretical models have been proposed to explain these experimental observations. However, while successful in certain aspects, these models only focus on average learning behaviors, and they cannot explain some crucial features of learning. For example, they cannot account for the dynamics of the whole distributions of the motor outputs and cannot explain why learning speed and magnitude negatively correlate with the perturbation size. Here we propose a model that includes multiple hidden dynamical variables, which collectively generate the desired motor command. The model is a multi-dimensional Bayesian filter that deals with the dynamics of non-Gaussian joint distributions of those hidden variables. Our model explains simultaneously the dynamics of distributions of the songbird vocal behaviors in various experiments, including: (i) adaptations after step changes or ramps in the error signal, and (ii) dynamics of relaxation following removal of the perturbation. We expect this model can be applied to data from other species and sensorimotor behaviors. |
Tuesday, March 5, 2019 3:06PM - 3:18PM |
H66.00004: Structure from noise: Mental errors yield abstract representations of events Christopher Lynn, Ari E Kahn, Danielle Bassett Humans are adept at uncovering complex associations in the world around them, yet the underlying mechanisms remain poorly understood. Intuitively, learning the higher-order structure of statistical relationships should involve sophisticated mental processes, expending valuable computational resources. Here we propose a competing perspective: that higher-order associations actually arise from natural errors in learning. Combining ideas from information theory and reinforcement learning, we derive a novel maximum entropy model of people’s internal expectations about the transition structures underlying sequences of ordered events. Importantly, our model analytically accounts for previously unexplained network effects on human expectations and quantitatively describes human reaction times in probabilistic sequential motor tasks. Additionally, our model asserts that human expectations should depend critically on the different topological scales in a transition network, a prediction that we subsequently test and validate in a novel experiment. Generally, our results highlight the important role of mental errors in shaping abstract representations, and directly inspire new physically-motivated models of human behavior. |
Tuesday, March 5, 2019 3:18PM - 3:30PM |
H66.00005: Can vocalizations predict mating pairs in a society of songbirds? A maximum-entropy Ising model approach Eve Armstrong, Clelia de Mulatier, David White, Marc Schmidt, Vijay Balasubramanian During mating season, most species of songbird engage in a societal evolution wherein monogamous pairs “freeze out”. Presumably, these bonds are a recipe for successful procreation. The means by which all individuals “agree” on this structure is unknown. The role of song is significant1, but its mechanism of orchestrating bonding is obscure. Moreover, a dynamical systems modeling approach would be premature, as it is not clear how to define the variables. |
Tuesday, March 5, 2019 3:30PM - 3:42PM |
H66.00006: Analytical corrections to entropy for under-sampled discrete distributions. Damian Hernandez, Ahmed Roman, Ilya Nemenman Estimating entropy of probability distributions of various data sets is a common question in modern data analysis. A common problem is that the number of independent samples obtained in experiments is limited, so that many states are under-sampled and naïve entropy estimators are inaccurate. Previous studies found that the statistics of states that occur in data sets multiple times (coincidences) provide useful corrections to entropy estimates in the extremely under-sampled regime. These corrections are largely numerical in nature and so provide little insight to which features of the dataset cause them. Here, we present analytical approximations to a coincidence-based entropy estimators, which shed some light on this question. |
Tuesday, March 5, 2019 3:42PM - 3:54PM |
H66.00007: Inferring geometric embeddings for single cell data Mor Nitzan, Nikos Karaiskos, Nir Friedman, Nikolaus Rajewsky Massively multiplexed sequencing of RNA in individual cells is transforming basic and clinical life sciences. However, in standard experiments sequenced cells do not retain information about their original spatial context although it is crucial for understanding cellular function. Recent attempts to overcome this fundamental problem rely on employing additional imaging data which can guide spatial mapping. Here we present a conceptually different approach that allows to reconstruct spatial positions of cells in a variety of tissues without using reference imaging data. We first show for several complex biological systems that distances of single cells in expression space monotonically increase with their distances across tissues. We therefore seek to map cells to tissue space such that this principle is optimally preserved, while incorporating imaging data when available. We show that this optimization problem can be cast as an optimal transport problem and solved efficiently. We apply our approach successfully to reconstruct the mammalian liver and intestinal epithelium as well as the fly embryo. Our results demonstrate a simple spatial expression organization principle that can be used to infer meaningful spatial position probabilities from the sequencing data alone. |
Tuesday, March 5, 2019 3:54PM - 4:06PM |
H66.00008: Interpreting time series data from dynamic signaling pathways in single cells Weerapat Pittayakanchit, Kabir Husain, Arvind Murugan A growing body of evidence suggests that cells encode information in the dynamics of signaling molecules. For instance, both the identity and dose of different external ligands may be encoded in the temporal dynamics of a single transcription factor. Understanding which aspects of experimental time-series are informative, and which can plausibly be decoded by cells given biochemical constraints, remains an open problem. Here, we combine modified versions of interpretable machine learning techniques, such as InfoGAN, with domain knowledge of the NF-kB pathway to obtain insights on how experimental time-series data of NF-kB encodes information about the ligands TNFa and IL-2. |
Tuesday, March 5, 2019 4:06PM - 4:18PM |
H66.00009: Learning image models for optimal information extraction: image registration from first principles Colin Clement, Matthew Bierbaum, James Patarasp Sethna We demonstrate an unbiased method of image registration which has errors consistent with the Cramer-Rao bound (CRB) by using Super Registration: learning an optimal model for the underlying image and shifting that tomatch the data. Image registration is the inference of transformations relating noisy and distorted images. Fundamental in computer vision, experimental physics, and medical imaging, even in the simplest case of translation, known methods are biased and none achieve the CRB. Cutting edge experiments operator at extreme limits of signal-to-noise, for example, low-dose TEM imaging of sensitive biological materials. It is in these high-noise scenarios when existing registration techniques fail to correctly infer shifts. Following Bayesian inference, we prove that the standard method of shifting one image to match another cannot reach the CRB. We reach the theoretical lower bound in shift resolution and extract a higher resolution, de-noised model of the latent image. Finally, while sub-pixel errors in shift inference do not dramatically change the reconstructed image for oversampled data, we show that using our new registration method can lead to 10× more precise particle tracking. |
Tuesday, March 5, 2019 4:18PM - 4:30PM |
H66.00010: Stochastic Nonlinear Dynamics of Confined Cell Migration David Brückner, Alexandra Fink, Christoph Schreiber, Joachim Rädler, Chase Broedersz In many biological phenomena, cells migrate through confining environments. To study such confined migration, we place migrating cells in two-state micropatterns, in which the cells stochastically migrate back and forth between two square adhesion sites connected by a thin bridge. We adopt a data-driven approach where we learn an equation of motion directly from the experimentally determined short time-scale dynamics, decomposing the migration into deterministic and stochastic contributions. This equation captures the dynamics of the confined cell and accurately predicts the transition rates between the sites. We thereby derive the emergent non-linear dynamics that governs the migration directly from experimental data. In particular, we find that the deterministic dynamics is poised near a bifurcation between a limit cycle and bistable behaviour. As a result, we find that cells are deterministically driven into the thin constriction; a process that is sped up by noise. Our approach yields a conceptual framework that may be extended to describe cell migration in more complex confining environments. |
Tuesday, March 5, 2019 4:30PM - 4:42PM |
H66.00011: Entropy-based idealization of complex single-molecule time trajectories from nanoelectronic biosensors: application to the detection and modeling of non-stationary molecular dynamics Mohamed OUQAMRA, Delphine Bouilly Single-molecule field-effect transistors (smFET) have been recently exploited to probe molecular dynamic events occurring at the single-molecule scale. Hidden Markov models (HMMs) are commonly used to model single-molecule kinetic trajectories, but they require restrictive assumptions and a priori knowledge of the likely kinetic model, both rarely met in real experiments. In particular, a major challenge in extracting kinetics from smFET data relies on the non-stationarity of the recorded signals due to noise and drifts. Here, we propose a new approach based on machine learning to retrieve the hidden trajectory between molecular states. Our method is a two-step algorithm based on compression of the raw data using a minimum description length cost function, followed by a k-medoid clustering of the compression patterns. A decision-aid tool automatically selects the multi-state model providing the best-fitting trajectory. Based on tests on synthetic and experimental data, we found that this entropy-based method allows to extract a robust idealized trajectory, without requiring any prior or supervision. We also found that using this idealized trajectory as a prior for HMM analysis provides better performances for the detection and modeling of non-stationary molecular dynamics. |
Tuesday, March 5, 2019 4:42PM - 4:54PM |
H66.00012: Predicting sequence-specific mutation rates in DNA Martha Villagran, Nikolaos Mitsakos, Ricardo Azevedo, John H Miller Mutations play a critical role in molecular evolution and the development of many diseases, such as cancer and neurodegenerative diseases. A growing body of evidence shows that rates of mutation in DNA are not only highly variable, but also influenced by the specific sites and nearby sequences. We have found that the physics of electron hole localization, most pronounced at or near guanine sites, plays a significant role in influencing sequence-specific mutation rates [M.Y. Suárez-Villagrán, R. B. R. Azevedo, & J. H Miller, Jr., Genome Biology & Evolution, 10, 1039 (2018)]. Most recently we have been applying the predictive capability of Deep Neural Network architectures, among other machine learning approaches, to predict and validate genetic mutation rates in human mitochondrial DNA. Given a segment of a sequence from the neighborhood of a specific base pair, we are able to predict mutation rates with much greater accuracy than that of a random predictor. We are currently testing the limits of automatic predictors for similar tasks, with an aim towards better understanding of both evolution and the emergence of somatic disease states, such as cancer. |
Tuesday, March 5, 2019 4:54PM - 5:06PM |
H66.00013: The Effects of Non-Specific Binding Kinetics on Fluorescence Activated Cell Sorting (FACS) Bhaven Mistry, Thomas Chou Fluorescence activated cell sorting (FACS) is extensively used in biological studies to differentiate cells of interest (mutants) from control cells (wild-types). For mutant cells characterized by expression of a distinct membrane surface structure, fluorescent marker probes can be designed to bind specifically to those structures, resulting in a sufficiently high fluorescence intensity that indicates a mutant cell. However, endogenous membrane structures on wild-type cells may nonspecifically bind to the probes, resulting in false positive results. These same endogenous membrane structures would also be present on mutant cells, allowing both specific and non-specific binding to a single cell. We create a kinetic model of fluorescent probe binding dynamics by tracking populations of mutant and wild-type cells with differing numbers of probes bound specifically and non-specifically. By assuming the suspension is in equilibrium prior to cytometry, we analytically derive a likelihood function of the FACS output in order to infer the total number of mutant cells while accounting for the non-specific binding of probes. We further show how our model can be used to infer unknown binding rates of fluorescent markers if cell counts are a priori known. |
Tuesday, March 5, 2019 5:06PM - 5:18PM |
H66.00014: Reverse-time inference of targeted biological dynamics Nicolas Lenner, Stephan Eule, Fred Wolf Mesoscopic bio-systems typically evolve towards functionally important target-states, such as cell cycle checkpoints or decision boundaries for the release of complex behavior. To infer the underlying directional out-of-equilibrium dynamics from such data, we develop a theory of target-state-aligned ensembles that reveals whether and when the system can be represented by a single, effective stochastic equation of motion. We show how, in this equation, genuine biological forces can be separated from spurious forces, which, invariably arise from target-state-alignment. We apply our inference scheme to canonical biological examples such as cell division and morphogenesis. |
Tuesday, March 5, 2019 5:18PM - 5:30PM |
H66.00015: Information flow and the accuracy of concentration measurements in a genetic network Marianne Bauer, William Bialek, Thomas Gregor, Mariela D Petkova, Eric Wieschaus Many genes are regulated by transcription factor (TF) proteins which bind to DNA and control the synthesis of messenger RNA (mRNA). The information which output mRNA or protein levels carry about input TF concentrations is a measure of regulatory power. All this information passes through a bottleneck, the occupancy of the relevant binding sites along the DNA, and random arrival of TFs at these sites sets an irreducible noise level, which in turn limits the information capacity. We explore these issues in the gap gene network of the early fly embryo, where recent work shows that the concentrations of these transcription factors carry enough information to specific cellular position to 1% precision along the length of the embryo. But how accurately would the system need to “measure” these concentrations in order to extract this information? We show that these measurements need to be more accurate than is plausible given the physical limits at a single binding site, suggesting that the complex array of multiple binding sites provides a solution to the problem of efficient information transmission. |
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