NITMB Seminar Series Recordings
Joseph Hibdon
Recorded on 11/21/2025
Title: Invasive Species to Bacteria Modeling: Examples of Interdisciplinary Biomathematics Research at a PUI
Abstract: There are many ways to engage students at a Primarily Undergraduate Institution (PUI). In this presentation, I will showcase two biology-driven examples. Both projects have involved numerous undergraduate and graduate students from across STEM departments at Northeastern Illinois University (NEIU). The first project focuses on modeling invasive species using dynamical systems, with the added twist that the invasive species cross-breeds with a native species to form a hybrid species. In the NEIU Biology Department, students have been investigating such cross-breeding in cattails in Northern Illinois. The mathematical model is relatively simple but provides students with opportunities to explore dynamical systems and develop numerical skills while examining potential outcomes of hybridization. The second project centers on bacterial modeling, specifically the pattern formation that emerges when Myxobacteria are grown under starvation conditions. Rather than using partial differential equations, we employ agent-based modeling in which each bacterium has distinct decision rules. This project has expanded to include the use of large language models to predict secondary metabolites produced by these bacteria and to evaluate the potential of these metabolites in combating cancer cells.
Jingyi Jessica Li
Recorded on 11/07/2025
Title: Statistical Learning in the Wild: Rethinking Discovery in the AI and Data Era
Abstract: Balancing false discovery rate (FDR) control with high statistical power remains a central challenge in high-dimensional variable selection. While several FDR-controlling methods have been proposed, many degrade the original data -- by adding knockoff variables or splitting the data -- which often leads to substantial power loss and hampers detection of true signals. We introduce Nullstrap, a novel framework that controls FDR without altering the original data. Nullstrap generates synthetic null data by fitting a null model under the global null hypothesis that no variables are important. It then applies the same estimation procedure in parallel to both the original and synthetic data. This parallel approach mirrors that of the classical likelihood ratio test, making Nullstrap its numerical analog. By adjusting the synthetic null coefficient estimates through a data-driven correction procedure, Nullstrap identifies important variables while controlling the FDR. We provide theoretical guarantees for asymptotic FDR control at any desired level and show that power converges to one in probability. Nullstrap is simple to implement and broadly applicable to high-dimensional linear models, generalized linear models, Cox models, and Gaussian graphical models. Simulations and real-data applications show that Nullstrap achieves robust FDR control and consistently outperforms leading methods in both power and efficiency.
Jingyi Jessica Li
Recorded on 11/07/2025
Title: Statistical Learning in the Wild: Rethinking Discovery in the AI and Data Era
Abstract: Balancing false discovery rate (FDR) control with high statistical power remains a central challenge in high-dimensional variable selection. While several FDR-controlling methods have been proposed, many degrade the original data -- by adding knockoff variables or splitting the data -- which often leads to substantial power loss and hampers detection of true signals. We introduce Nullstrap, a novel framework that controls FDR without altering the original data. Nullstrap generates synthetic null data by fitting a null model under the global null hypothesis that no variables are important. It then applies the same estimation procedure in parallel to both the original and synthetic data. This parallel approach mirrors that of the classical likelihood ratio test, making Nullstrap its numerical analog. By adjusting the synthetic null coefficient estimates through a data-driven correction procedure, Nullstrap identifies important variables while controlling the FDR. We provide theoretical guarantees for asymptotic FDR control at any desired level and show that power converges to one in probability. Nullstrap is simple to implement and broadly applicable to high-dimensional linear models, generalized linear models, Cox models, and Gaussian graphical models. Simulations and real-data applications show that Nullstrap achieves robust FDR control and consistently outperforms leading methods in both power and efficiency.
Samuel Muscinelli
Recorded on 11/14/2025
Title: A Network Mechanism for Perceptual Learning
Abstract: Organisms continually tune their perceptual systems to the features they encounter in their environment. We have studied how this experience reorganizes the synaptic connectivity of neurons in the olfactory (piriform) cortex of the mouse. We developed an approach to measure synaptic connectivity in vivo, training a deep convolutional network to reliably identify monosynaptic connections from the spike-time cross-correlograms of 4.4 million single-unit pairs. This revealed that excitatory piriform neurons that respond similarly to each other are more likely to be connected. We asked whether this like-to-like connectivity was modified by experience but found no effect. Instead, we found a pronounced effect of experience on the connectivity of inhibitory interneurons. Following repeated encounters with a set of odorants, inhibitory neurons that responded differentially to these stimuli both received and formed a high degree of synaptic connections with the cortical network. The experience-dependent organization of inhibitory neuron connectivity was independent of the tuning of either their pre- or their postsynaptic partners. These results suggest the existence of a cell-intrinsic, non-Hebbian plasticity mechanism that depends only on the odor tuning of the inhibitory interneuron. A computational model of this plasticity mechanism predicts that it increases the dimensionality of the entire network’s responses to familiar stimuli, thereby enhancing their discriminability. We confirmed that this network-level property is present in physiological measurements, which showed increased dimensionality and separability of the evoked responses to familiar versus novel odorants. Thus a cell-intrinsic plasticity mechanism acting on inhibitory interneurons may implement a key component of perceptual learning: enhancing an organism’s discrimination of the features particular to its environment.
Cody Fitzgerald
Recorded on 10/24/2025
Title: Data-driven modeling of complex biological systems
Abstract: Finding simple data-driven and interpretable models of complex biological systems requires engaging with experimental and real-world data, which often come with a range of challenges. Typical biological data sets often feature sources of extrinsic and intrinsic noise, infrequent temporal sampling of the dynamics, and, usually, not all relevant states of the system can be experimentally measured. Additionally, for many biological systems, the true dimensionality is unknown. In this talk, I walk through a series of recent modeling projects using experimental and real-world data, each featuring a different data-related challenge. In particular, I will discuss using wolves as an ecological control strategy to limit the spread of Chronic Wasting Disease in deer populations, an inference problem in which we fit 13,824 separate gene regulatory network models to an RNA-seq data set in an attempt to learn what can honestly be said about the underlying regulatory mechanism, and using model selection for partially observed systems to infer a physiological model of the thermoregulatory system for Arctic hibernators. Finally, I will describe recent ongoing work using a super high-dimensional spatio-temporal data set from the City of Chicago, in which we attempt to understand the city as a cellular organism and infer its "organelles" directly from data.
Siwei Wang
Recorded on 10/17/2025
Title: Search for common principles for efficient representation in both neural coding and AI
Abstract: In the wild, survival often depends on anticipating what happens next. A fly must escape a predator's strike before its visual system even finishes processing the threat. A salamander needs to track prey moving through completely different environments as it migrates from aquatic to terrestrial environments. How neural systems extract actionable insights within a split-second while neural processing is inherently laggy? Given this temporal constraint, predictive information is one such common computation. By encoding patterns predictive of future maneuvers, specialized motion sensitive neurons in the fly visual system enables them to steer animals away from visual threat within 40 ms, half the time as our eye blinks. Similarly, in the salamander retina, the encoding of predictive information sculpts the population code to establish a low-dimensional, transferable representation that seamlessly generalizes across radically different natural scenes. In addition, when this representation combines static and dynamic features synergistically to anticipate future states, its compression rivals the most state-of-the-art video compression algorithms. The above insights from biology inspires new theories in AI. By extracting the general principles for the above biological efficient representations, the newest work from my group demonstrates that such principle also applies to feature engineering in contrastive learning writ large. The overarching goal of this walk is to establish a virtuous cycle where developing biologically plausible AI systems helps interrogate unknown encoding mechanisms in the brain, while these discovered neural principles simultaneously inspire more efficient computational architecture.
Richard Carthew
Recorded on 10/10/2025
Title: Emergence of a hexagonal lattice of photoreceptor neurons by tissue-scale mechanics
Abstract: Animal development is marked by a progressive emergence of ordered structures from disordered precursors. Various models are used to describe such developmental phenomena, and a long-standing model has been the Reaction-Diffusion Model. One such RD model nicely describes the emergence of a hexagonal lattice of photoreceptor neurons in the developing eye of the fruit fly Drosophila. The model relies on activator-inhibitor pairs that are essential for the patterning in vivo. Nevertheless, the RD model has not been tested by real-time observation of the developmental dynamics. We did so using live-imaging microscopy and discovered a completely unexpected mechanical component in eye development. Ordered periodic flows of cells within the eye epithelium create the hexagonal lattice from a pseudo-square lattice prepattern of greater disorder. Thus, an RD mechanism, if involved, is likely generating the prepattern, which then evolves by a mechanical mechanism into the mature hexagonal lattice of the compound eye.
Sergei Maslov
Recorded on 10/03/2025
Title: Crossfeeding Dynamics in Energy-Limited and Auxotrophic Systems
Abstract: Microorganisms in many environments survive by sharing nutrients with each other in a process called crossfeeding. Some microbes are extremely slow-growing (taking weeks or years to divide instead of hours), while others are auxotrophs that have lost the ability to make essential nutrients and must obtain them from neighboring species. I will describe two distinct modeling approaches for understanding crossfeeding: (1) a thermodynamic consumer-resource model for slow-growing communities in energy-limited environments (George, Wang & Maslov, ISME J, 2023) and (2) a higher-order interaction model for auxotrophic communities dependent on essential resource exchange (Wang & Maslov, Cell Systems, in press). The thermodynamic model is applicable to slow-growing communities in which metabolic byproducts from reactions close to thermodynamic equilibrium create crossfeeding networks. We derive the principle of maximum free energy dissipation in this model and demonstrate how functional convergence emerges despite taxonomic differences in crossfeeding partnerships. The model successfully predicts functional convergence in anaerobic digesters, showing how crossfeeding networks stabilize at low dilution rates. The auxotroph model examines communities where organisms are genetically unable to synthesize essential nutrients (e.g. amino acids) and must rely on crossfeeding from other species. Using graphical and algebraic methods, our approach reveals how amino acid crossfeeding creates higher-order interaction networks that enhance community resilience to environmental fluctuations through metabolic complementarity. Applied to experimental data from synthetic E. coli communities, the model accurately predicted the survival of 3 out of 4 strains in a 14-member auxotroph community, correctly identifying key mutualistic crossfeeding partnerships.
Arjun Raman
Recorded on 09/19/2025
Title: Towards a generative theory of emergence
Abstract: Complex systems that emerge naturally comprise many parts that interact in unintuitive manners and are therefore perceived as difficult to understand. However, such systems share structural qualities that enable comparison, thereby suggesting the possibility of generative principles to be gleaned. Current learning frameworks (i.e. deep-learning) approximate data distributions without considering the anatomy of the generative process. Here, we show that inferring just two constraints across the extant diversity of complex systems is sufficient to learn generative logic. The first constraint defines entropic scales of organization: the hierarchical distribution of information across an ensemble of related systems. The second constraint is the sequential joining of entropic scales: lower entropy scales anchor variability within higher-entropy scales. Using this framework, we created a new class of generative models—Layered Restricted Boltzmann Machines (LRBMs)—for several systems: images, language, and enzyme design. LRBMs were remarkably lightweight computationally, inherently interpretable, and lay the framework for rational engineering towards a desired phenotype. Taken together, our results suggest that complex structured systems might arise from a hierarchy of entropic scales that sequentially anchor variation. This quality provides a blueprint for generation across natural and artificial domains.
Leor Weinberger
Recorded on 5/30/2025
Title: Therapeutic Interfering Particles
Abstract: Theories have long proposed that Therapeutic Interfering Particles (TIPs)—a class of Defective Interfering Particles (DIPs) engineered to have R0>1 and stably propagate to new cells—could overcome major barriers to disease control. TIPs exert their antiviral effects through competition for viral replication and packaging machinery and as a consequence of their conditional replication were predicted to act as single-administration, resistance-proof antivirals (Metzger et al., 2011). We recently reported engineering of the first TIPs and the first demonstration of the hypothesized TIP therapeutic effect (Tanner et al. 2019; Chaturvedi et al. Cell 2021; Notton et al. 2021; Chaturvedi et al. PNAS, 2022). For SARS-CoV-2, a single, intranasal administration of TIPs to Syrian golden hamsters dramatically lowers virus by 3-Log10s and protects against severe disease, even for neutralization-resistant variants (beta, delta, etc…). For HIV-1, a single injection of TIPs durably suppresses highly-pathogenic virus, long-term, by 4-Log10s, in non-human primates—without emergence of resistance—and protects against death (Pitchai et al. Science 2024). These data provide proof-of-concept for a novel class of antivirals with the capacity to overcome key challenges in infectious disease control.
Vincenzo Vitelli
Recorded on 05/23/2025
Title: Learning dynamical behaviors with minimal ingredients
David Freedman
Recorded on 5/16/2025
Title: Brain Circuits and Computations of Flexible Decision Making
Abstract: Humans and other animals are adept at learning to perform cognitively-demanding behavioral tasks. Neurophysiological recordings in non-human primates during such tasks find that task-related cognitive variables are encoded across a wide network of brain regions, which transform sensory encoding of the external world into decisions and actions. This talk will discuss a line work which employs large scale recordings from neuronal populations in the primate brain during visually-based decision making, as well as parallel AI-based computational modeling approaches to generate hypotheses and predictions for further analyses and experiments.
Krishna Shrinivas
Recorded on 05/02/2025
Title: Rational design of fuzzy biomolecular ensembles
Abstract: Recent advances in machine learning have accelerated the design of folded proteins. However, many proteins and protein regions lack stable folds and are intrinsically disordered (IDPs). The plastic and shape-shifting conformational ensemble that the sequence of an IDP encodes for dictates biological function. In this talk, I will discuss methods we are developing to advance the design of IDPs through strategies to model and invert the sequence-ensemble-function paradigm.
Elizabeth Jerison
Recorded on 4/11/2025
Title: Investigating the internal dynamics of inflammatory response
Abstract: Our immune system detects and responds rapidly to unexpected challenges from invading pathogens. These responses are orchestrated by a variety of cells throughout the body, and the molecules that they use to communicate, creating rich internal dynamics. These include feed forward interactions that are crucial to system function, but can be dangerous to the host. While we have a nearly complete list of the components of this complex system, understanding the molecular and mathematical drivers of its dynamics remains a vast challenge. I will discuss recent progress on two aspects of this problem. First, I will discuss steps towards disentangling the role of tissue, as opposed to the external stimulus, in determining spatiotemporal patterns of gene expression during an inflammatory response in a model system. Second, I will discuss a new computational method that uses contrastive learning to address the problem of characterizing dynamical behavior when there is no clear a priori model, a scenario that often arises in immune contexts. I will conclude by discussing the avenues that these open for investigating how the collective action of cells and molecules within tissues control inflammatory responses.
Luis Bettencourt
Recorded on 03/07/2025
Title: Redefining Fitness: Evolution as a Dynamic Learning Process
Abstract: Evolution is the process of optimal adaptation of biological populations to their living environments. This is expressed via the concept of fitness, defined as relative reproductive success. However, it has been pointed out that this procedure is incomplete and logically circular. To address this issue, several authors have called for new ways to re-define fitness through the relationship between phenotypes and their environment, but a general approach has remained elusive. In this talk, I show that fitness, defined as the likelihood function that follows from mapping population dynamics to Bayesian learning, provides a principled solution to this problem. I will show how probabilistic models of fitness can easily be constructed in this way, and how their averages acquire meaning as information. I will also show how this approach leads to powerful tools to analyze problems of evolution in stochastic environments, game theory, and selection in group-structured populations.
Matthias Steinrücken
Recorded on 1/24/2025
Title: Inferring selection from time-series genetic data
Abstract: Detecting and quantifying the strength of selection is a main objective in population genetics. Since selection acts over multiple generations, many approaches have been developed to detect and quantify selection using genetic data sampled at multiple points in time. Such time series genetic data is commonly analyzed using Hidden Markov Models, but in most cases, under the assumption of additive selection. However, many examples of genetic variation exhibiting non-additive mechanisms exist, making it crucial to develop methods that can characterize selection in more general scenarios. In this talk, I present approaches for the inference of general diploid selection coefficients, where the heterozygote and homozygote fitness are parameterized independently. I furthermore introduce a framework to identify bespoke modes of diploid selection from given data, as well as a procedure for aggregating data across linked loci to increase power and robustness. Extensive simulation studies show that these methods accurately and efficiently estimates selection coefficients for different modes of diploid selection across a wide range of scenarios. Applying the approaches to ancient DNA samples from Great Britain in the last 4,450 years, shows evidence for selection in several genomic regions, including the well-characterized LCT locus.
Jasmine Nirody
Recorded on 11/15/2024
Title: A tale of two motilities: adaptive locomotion in complex, changing environments
Abstract: I will discuss two (quite different!) adaptive locomotive modes that are able to contend with the heterogeneity and uncertainty inherent to natural environments: flagellated swimming in bacteria and walking in panarthropods. (1) Bacteria have evolved mechanisms at both the molecular and whole-cell level to facilitate motility through a wide range of complex environments. At the molecular level, we characterize mechanosensitivity in the bacterial flagellar motor (BFM), the rotary nanomachine that propels bacterial swimming. At the whole cell level, we explore how bacterial motility patterns adapt to confined, disordered environments. We discuss the features of these results that generalize across bacterial species (and those that don't!), underscoring the importance of considering flagellar motility in the context of a bacterium's environment and evolutionary history. (2) Panarthropods are a diverse clade containing insects, crustaceans, myriapods, and tardigrades. We show that inter-limb coordination patterns in freely-behaving tardigrades replicate several key features of walking in insects across a range of speeds and substrates. In light of these functional similarities, we propose a simple universal locomotor circuit capable of robust multi-legged control across body sizes, skeletal structures, and habitats.
Veronica Ciocanel
Recorded on 11/08/2024
Title: Parameter identifiability for PDE models of fluorescence microscopy experiments
Abstract: The dynamics of intracellular proteins is key to many cellular functions. One of the most versatile experimental techniques for probing protein dynamics in living cells is FRAP (fluorescence recovery after photobleaching). This experiment generates time-series data that average out spatial information about diffusion, transport, and binding dynamics of proteins. Partial differential equations (PDE) models provide the appropriate framework to model the fluorescence dynamics and to infer parameters such as diffusion coefficients or reaction rates. However, it is not known whether these parameters can be identified based on the spatially-averaged data available from FRAP experiments. We recently investigated limitations of known methods in assessing parameter identifiability for PDE models and proposed methods for learning parameter combinations based on re-parametrization and profile likelihoods analysis. In this work, our motivation stems from studying dynamic RNA binding proteins in biomolecular condensates that play key roles in the development of frog oocytes.
Jorge Nocedal
Recorded on 10/25/2024
Title: How is it Possible to Train Deep Neural Networks?
Abstract: In 1961, Minsky, one of the founders of AI, perceived a fundamental flaw within the burgeoning field of artificial neural networks. He doubted that such a nonlinear system could be effectively trained using gradient methods, because unless the “structure of the search space is special, the optimization may do more harm than good.” Fast forward to today, and we observe deep neural networks — far more complex than those envisioned at the field's inception — being successfully trained with methods akin to gradient descent. It has, indeed, become evident that the objective function displays a highly benign structure that we are only starting to comprehend. In this lecture, I aim to summarize our current understanding of this enigmatic optimization process. I will discuss several themes, including intrinsic dimensionality, the optimization landscape, and implicit regularization, all within the context of deep networks and large language models. Speaker Bio: Jorge Nocedal is the Walter P. Murphy Professor of Industrial Engineering and Management Sciences and (by courtesy) Engineering Sciences and Applied Mathematics at Northwestern University. Nocedal is also the Director of the Center for Optimization and Statistical Learning. Nocedal's main area of research is optimization, with applications in machine learning, engineering design, and the physical sciences. Research activities range from the design of new algorithms, to their software implementation and mathematical analysis. Areas of emphasis include large scale problems (with millions of variables), optimization under uncertainty, and parallel computing.
Thomas P Wytock
Recorded on 10/11/2024
Title: Cell reprogramming and bacterial memory predicted by mathematical modeling and transfer learning
Abstract: Determining how molecular changes propagate to affect the whole cell is a fundamental goal in systems biology. The challenge inherent to this pursuit is the number and diversity of molecular entities that give rise to cell behavior. In this talk, I will show how nonlinear dynamics and machine learning uncover unexpected behaviors in recent studies concerning irreversibility in bacterial regulatory networks [1] and cell reprogramming [2]. In the first study, Boolean networks reveal that the same DNA sequence can counterintuitively encode multiple behaviors through information stored in the state of the gene regulatory network. In the second study, a machine learning model provides an effective representation of the regulatory interactions between genes and cell phenotypes. We use this model to identify gene perturbations that can reprogram cells to different phenotypes. Both contributions demonstrate how the abundance of biological data provides opportunities for mathematical modeling to derive unexpected insights. Speaker Bio: Thomas P. Wytock is a postdoctoral researcher in the Motter Group at Northwestern University, where he focuses on the modeling of genetic networks.