Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session A08: Network Theory and Application to Complex Systems IFocus Recordings Available
|
Hide Abstracts |
Sponsoring Units: GSNP Chair: Filippo Radicchi, Indiana University Room: McCormick Place W-179B |
Monday, March 14, 2022 8:00AM - 8:36AM |
A08.00001: System-state dynamics and recurrence of temporal networks Invited Speaker: Naoki Masuda Various empirical networks are regarded to vary over time (i.e., temporal networks) on a timescale of relevant dynamics (e.g., epidemic dynamics) occurring on them. Temporal network data are complex because the structure of the network at each time point is quite often complex already, and such a complex network evolves over time. We propose two methods based on stochastic processes and nonlinear time series analysis to simplify temporal networks with the aim of capturing dynamics of temporal network data. In the first approach, we reduce dynamics of networks into that of a single "system state" that switches over time among a relatively small number of possible states. For example, in temporal contact network data measured in a primary school, we find that one of the two inferred states corresponds to class time and the other state corresponds to lunch time. In the second approach, we extend recurrent plots and recurrent quantification analysis, which is a nonlinear time series analysis method proposed in 1980s in physics community, to the case of temporal networks. Specifically, we collect all instances of recurrence in the sense that the network measured at time t_1 is similar to that at t_2. This information allows us to draw recurrence plots, which one can further quantify. I will demonstrate this method with neural data recorded from individuals with epilepsy and with temporal airport networks. With both methods, we crucially need to measure distances between networks at pairs of time points and use that information to create overviews of temporal networks. |
Monday, March 14, 2022 8:36AM - 8:48AM |
A08.00002: Template-Free Reaction Networks Enable Predictive and Automated Analysis of Complex Electrochemical Reaction Cascades Samuel M Blau, Daniel Barter, Evan Spotte-Smith Chemical reaction networks (CRNs) are powerful tools for obtaining mechanistic insight into complex reactive processes. However, existing techniques rely heavily on chemical intuition and prior knowledge, limiting their applicability in domains where reaction mechanisms or products are unknown and where potential energy surface exploration is computationally intractable. Here we report new methods of CRN generation and analysis that overcome these limitations. By constructing CRNs using filters, rather than templates, we preserve species and reactions that are unintuitive but fundamentally reasonable. The resulting massive CRNs can then be interrogated via stochastic methods, revealing thermodynamically bounded reaction pathways to species of interest and automatically identifying network products. We apply this methodology to study solid electrolyte interphase (SEI) formation in Li-ion batteries. Our methods automatically recover SEI products from the literature and predict previously unknown species; the predicted formation mechanisms to select products are then validated using first-principles calculations. This methodology enables the efficient de novo exploration of vast chemical spaces, with the potential for diverse applications across thermochemistry, electrochemistry, and photochemistry. |
Monday, March 14, 2022 8:48AM - 9:00AM |
A08.00003: Towards dismantling healing illicit & counterfeit medicines seller networks (ICMSN) using percolation theory & machine learning: A simulation study Timothy A Burt, Ravi Sundaram, Nikos Passas, Mansoor Amiji, Muhammad Zaman, Ioannis A Kakadiaris Illicit and counterfeit medicines (ICM) are responsible for over half a million deaths annually, accounting for approximately $75B of the $962B global pharmaceutical market. To reduce the societal harm that comes from these products, new approaches which can identify, intervene, and disrupt the trade of ICM are needed. Meanwhile, recent work has demonstrated the potential of machine learning (ML) for dismantling complex societal networks of interest better than current SOTA analytical methods. This project takes a rigorous scientific approach to learn what makes ICM seller networks (ICMSN) so adaptive and resilient by finding the fundamental static building blocks of ICMSN and modeling flows of drugs and money on dynamic ICMSN. We discuss the network characteristics of a static ICMSN constructed in-house from actual data, finding that its disconnected structure can be clustered using community detection techniques. We adapt the Graph Dismantling with Machine Learning (GDM) framework to explore dismantling clusters of ICMSN, using percolation theory as an oracle for understanding critical and emergent phenomena on these networks. We also present various modeling strategies for healing & rewiring between and within ICMSN clusters and their effects on the outcome of GDM. |
Monday, March 14, 2022 9:00AM - 9:12AM |
A08.00004: The statistical physics of ranking and partial orders George Cantwell, Cristopher Moore Ranking things is a common and natural task that is performed across a wide variety of contexts, from sports fandom to scheduling. More generally, ranking is simply one form of ordering, i.e. finding permutations of objects that satisfy constraints. Other than ranking, ordering tasks include reconstructing the temporal dynamics of disease propagation from contact traces, or fitting network growth models. On the basis of partial information, these tasks are computationally complex – #P-hard in the worst case, making them even harder than NP-complete problems. To make matters worse, real data is usually noisy: some observations are “wrong”, although we don’t know which. |
Monday, March 14, 2022 9:12AM - 9:24AM |
A08.00005: Optimization of the mean first passage time in complex networks Georgios Gounaris, Eleni Katifori Complex, dynamical processes in nature can be represented by networks, such as the stochastic jumps between numerous metastable states in the energy landscape of proteins, or the intracellular transport of molecules in the endoplasmic reticulum in eukaryotic cells. The state of the system can be represented as a random walk on a complex network, whose structure frequently combines both hierarchy and modularity. The graph topology and the link weight distribution affect the dynamical transitions between the states of the random walkers and ultimetely the trapping efficiency. To study this, we consider a complete graph with weighted links in which selected nodes correspond to absorbing states. We use an electrical circuit analogue to calculate the mean first passage time of random walkers to reach a trap. We propose an optimization rule according to which the edge weights remodel in order to minimize the mean first passage time and the cost to maintain the graph. We investigate how time varying, correlated traps, which can be activated and deactivated in a collective manner but with different correlations, can give rise on hierarchy and modularity in the network. |
Monday, March 14, 2022 9:24AM - 9:36AM Withdrawn |
A08.00006: Characterizing spatial networks using β-skeletons Szabolcs Horvát, Carl D Modes Most classic graph measures used by network science were designed to be applicable to arbitrary, generic graphs. However, the nodes of some real-world networks exist in physical space, with only nearby nodes being connected. This strongly constrains their possible connectivity structures, which renders many classic graph measures uninformative, and of limited use for classification. This is even more true in networks where only direct spatial neighbours are connected, and long-range connections are completely missing. Examples include various transport networks in biological organisms (such as vasculature), networks of streets, or fungal networks. In all these cases, node locations almost completely determine connectivity. We propose a novel approach to characterizing such networks through the concept of β-skeletons, a family of parametrized proximity graphs that captures spatial neighbour relations very well. By constructing a sequence of β-skeletons for the node locations of empirical spatial networks, we can characterize the local geometry of their node arrangements, and study how this influences their network structure. We demonstrate the method on several biological datasets. |
Monday, March 14, 2022 9:36AM - 9:48AM |
A08.00007: Evolution of modularity in biological networks Saul Huitzil, Cristian L Huepe At each scale of organization, living systems are composed of subsets of elements that interact closely with each other, forming modules that behave in turn as more complex elements of systems at higher levels of organization. For example, cells can be viewed as modular components that self-organize into organisms, and organisms as modules that interact to form ecologies. Understanding the underlying processes that can produce such modularity could help explain why evolution tends to form increasingly complex structures and dynamics. So far, however, we know very little about the origins of modularity in living systems. |
Monday, March 14, 2022 9:48AM - 10:00AM |
A08.00008: Computational inference of synaptic polarities in neuronal networks Istvan Kovacs, Thomas P Wytock, Michael Harris Synaptic polarity, i.e. whether synapses are inhibitory (-) or excitatory (+), is challenging to map, although being a key to understand brain function. Here, synaptic polarity is inferred with high precision considering three experimental scenarios. First, using the C. elegans connectome as an example, detailed neurotransmitter (NT) and receptor (R) gene expression is assumed. Such existing datasets are linked by the Connectome Model (CM), using a wiring rule network summarizing how NT-R pairs govern synaptic polarity, resolving 356 synaptic polarities in addition to the 1,752 known polarities. Second, known synaptic polarities are considered as an input, in addition to the NT and R expression data. Even without any wiring rules as an input, the Spatial Connectome Model (SCM) recovers 72% of the CM-resolved pairs at a threshold corresponding to >95% precision, while also inferring 118 of the remaining unknown polarities. Last, when no genetic information is available, the generalized Connectome Model (GCM) is introduced and compared to signed generalizations of network-based link prediction methods to infer synaptic polarities. Our results address current challenges in unveiling large-scale synaptic polarities, an essential step towards more realistic dynamical brain models. |
Monday, March 14, 2022 10:00AM - 10:12AM |
A08.00009: Application of network techniques to image analysis and outlier detection Cristina Masoller Recent advances in computer science and machine learning (ML) have generated highly efficient and unsupervised algorithms for the analysis of biomedical images, enabling cost-effective remote early detection of diseases. In this talk I will present various methods for ophthalmic image analysis, which use ML and network analysis. First, I will present a ML algorithm for the analysis of optical coherence tomography (OCT) images, which extracts features that discriminate between healthy and unhealthy subjects. Then, I will show that network analysis applied to the tree-like structure of the network of vessels in the retina returns features that discriminate between healthy subjects and those with glaucoma or diabetic retinopathy. Finally, I will discuss how the network percolation transition can be used for mining outliers in wide range of high-dimensional data sets, including ophthalmic images. |
Monday, March 14, 2022 10:12AM - 10:24AM |
A08.00010: Maximally robust neutral networks in the correlated phase of input-output maps Vaibhav Mohanty, Ard A Louis Systems which accept a sequence-based input and produce a nontrivial output appear widely across scientific disciplines. Examples include protein/RNA primary sequences mapping to their folded structures, gene regulatory network interactions mapping to expression cycles, and the set of interactions in a spin glass mapping to the ground state(s), among others. Naturally observed systems exhibit substantially higher robustness than would be expected from a random mapping of input to output. Casting the input-output map as vertex labeling of a Hamming graph, the robustness of an output to input mutations becomes a network-theoretic property. Using maximum entropy and a single constraint on global robustness, we present a general statistical physics model for discrete input-output maps which shows a universal transition between correlated (robust) and uncorrelated (fragile) phases. We analytically derive the scaling laws for robustness observed in natural systems and show that our model numerically reproduces other network topological features of natural input-output maps. We also elucidate the properties of subnetworks in the robust phase—maximally robust neutral networks known as "bricklayer's graphs"— including the robustness, which is related to the sums-of-digits function. |
Monday, March 14, 2022 10:24AM - 10:36AM |
A08.00011: Econophysics on networks Miron Kaufman Krugman [“The Self-Organizing Economy“, Blackwell, 1996] proposed a continuous model, in time and space, for the emergence of polycentric urban areas in the regional space. By using a discrete version of the Krugman model on spatial networks, we predicted the distribution of jobs among the different localities inside several economic regions in Ohio and Texas [S. Kaufman, M. Kaufman, M. Salling, Applied Network Science, 2019]. The model predicts new spatial distribution of jobs that emerges from the current fractions of jobs at time t and location x: nt,x through interactions among localities in a region. The market potential function of any location x at time t is Pt,x = Σy qx,ynt,y . It is determined by a network matrix qx,y connecting any two locations in the region. Employment gradually moves towards locations considered relatively attractive if their market potential is above the spatial average: Pt,x > <Pt >. Similarly, jobs move away from locations with below-average market potential. I will discuss a couple of general properties of the model. First, I will determine the stationary distribution and its stability. Second, I will show that the market potential, which is analogous to the fitting function, satisfies the Fisher [“The genetical theory of natural selection”, Oxford, 1930] equation of natural selection: <Pt+1> - <Pt> = var(Pt,x). |
Monday, March 14, 2022 10:36AM - 10:48AM |
A08.00012: Cliques influence on the spectrum of normal modes vibration in complex networks Oscar I Torres-Mena, Francisco J Sevilla-P'erez The collective vibrations in systems of linearly coupled oscillators has served as a simple paradigmatic model of collective phenomena. It is of particular recent interest, the case when the arrangement of the oscillators is not periodic, as is in the context of lattice vibrations of a crystalline solid, but when they lie on a complex network. In this work, the influence of the network-cliques number on the collective vibrations of linearly coupled oscillators is investigated for particular network topologies. Namely, we start with fully connected tree networks (no cliques) and different degree distributions. Then we systematically add higher order cliques to these trees to analyze their effect on the frequency spectrum. |
Monday, March 14, 2022 10:48AM - 11:00AM |
A08.00013: Statistical Physics of Associative Memory on Small World Networks Yash Gurbani, Santosh Kumar, Syed M Kamil Learning and associative memory are understood as emergent phenomena resulting from interactions between a complex network of neurons. It is well known that the structure of such a neural network heavily influences its function. Biological networks (e.g. neuronal network of the worm Caenorhabditis elegans) have been shown to exhibit small-world characteristics. To investigate the structure-function relationship in small-world networks, we simulate the Hopfield model of associative memory which uses the well-known spin-glass hamiltonian in statistical physics to model the neural network. We obtain estimates of memory capacity on a regular and a Watts-Strogatz (WS) network through numerical simulations. Further, we study how changing the probability of rewiring and local connectivity in a WS network affects the performance of associative memory. We find that the performance on small-world networks is as robust as that on random networks despite using only a fraction of connections, making the former biologically favorable. Our simulations are in agreement with experimental evidence found in the existing literature on small-world characteristics in biological networks and give deeper insights into this phenomenon. |
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