Embedded Files
1 – Quantum Operator-Based Real Autoencoder (QOBRA): A Quantum Autoencoder Algorithm for De Novo Molecular Design
Yue Yu, Francesco Calcagno, Haote Li, Victor S. Batista
Yale University, University of Bologna
We introduce a variational quantum autoencoder tailored for de novo molecular design named QOBRA (Quantum Operator-Based Real-Amplitude autoencoder). QOBRA leverages quantum circuits for real-amplitude encoding and the SWAP test to estimate reconstruction and latent-space regularization errors during back-propagation. Adjoint encoder and decoder operators enable unitary transformations and a generative process that ensures accurate reconstruction as well as novelty, uniqueness, and validity of the generated samples. We showcase the capabilities of QOBRA as applied to de novo design of Ca2+, Mg2+, and Zn2+-binding metalloproteins after training the generative model with a modest dataset.
2 ‐ VivariumEcoli: A Modular Framework for Predictive Whole-Cell Modeling
Neyamat Khan Tanvir1, Eran Agmon2
1. Systems Biology Graduate program, University of Connecticut Health Center, Farmington, CT, USA 06030
2. Center for Cell Analysis and Modeling, UConn Health, Farmington, CT, USA 06030
Predictive, mechanistic models of whole-cell activity are critical for understanding how E. coli combines several regulatory, metabolic, and signaling modules to survive and adapt in dynamic environment. Here, we introduce vEcoli, a modular, dynamical framework that integrates genome-scale metabolism with kinetic sub-models of transcriptional, post-transcriptional, translational, and post-translational control. In vEcoli, basic biological operations are encapsulated as discrete 'processes' (e.g., metabolism, transcription, translation, replication, and division), which interact via shared 'stores' that hold molecular state. Composites allow for hierarchical model creation, while workflows and experiments facilitate repeatable configuration, large-scale parameter sweeps. vEcoli's simulation engine combines constraint-based flux analysis of genome-scale metabolism with kinetic sub-models of macromolecular synthesis and regulatory networks, capturing both steady-state growth phenotypes and dynamic stress responses like nutrient shifts or antibiotic perturbations. Initial applications show accurate recapitulation of growth rates, metabolite pools, and proteome allocation under a variety of circumstances. Whole-cell modeling holds the promise of uniting diverse cellular processes—from metabolism and gene expression to regulatory circuits and division—into a single predictive framework. This framework paves the way for collaborative model development, synthetic‐biology design, and future extensions toward multi‐species and host–microbe interaction models.
3 – Astral architecture can enhance mechanical strength of cytoskeletal networks by modulating percolation thresholds
Brady Berg1, Jun Allard1,2
1. Mathematical, Computational & Systems Biology, UCI
2. Department of Mathematics, Department of Physics & Astronomy, Center for Complex Biological Systems, UCI
A repeated pattern in cytoskeletal architecture is the aster, in which a number of F-actin filaments emerge star-shaped from a central node. Aster-based structures occur in cytoplasmic actin, the early stages of the cytokinetic ring in yeast, and in the context of biomimetic materials engineering. In this work, we use computational simulation to show that there is an optimal number of filaments per aster that maximizes rigidity, even at a fixed density of F-actin. This nonlinear dependence holds for both the shear and extensional moduli. At physiological parameters, the maximum corresponds approximately to the same filaments-per-aster observed in recent super-resolution images of cortical F-actin. Furthermore, we find that increasing filaments-per-aster leads to dramatic increases in the sample-to-sample variability in network rigidity. We explain both effects using percolation theory, wherein the probability that a given network is productively connected exhibits a sharp dependence on parameters. The dependence of network rigidity on this nanoscale architectural feature may suggest a mechanism by which cells tune the physical properties of their actin networks locally and rapidly (since no new F-actin must be assembled) and may inform efforts to create adaptive synthetic metamaterials inspired by actin networks.
4 – Elucidating the role of ELAC-2 in regulating mitochondrial function via a novel anterograde response
Bharat Vivan Thapa, James Held, Chloe Hecht, Maulik Patel
Department of Biological Sciences, Vanderbilt University
Mitochondria are semiautonomous organelles essential for energy production, macromolecule biosynthesis, and signaling. Given these diverse functions, efficient communication between mitochondria and the nucleus is essential for maintaining cellular homeostasis. Eukaryotic cells have evolved elaborate crosstalk between the two organelles, broadly categorized into retrograde (mitochondria to nucleus) and anterograde (nucleus to mitochondria) signaling to ensure optimal function. Retrograde signaling, which consists of adaptive responses such as mitochondrial unfolded protein response (UPRmt), is well-characterized. In contrast, anterograde signaling involving preventative responses remains comparatively underexplored. We have uncovered a novel anterograde response involving ELAC-2, a tRNA processing enzyme, in regulating mitochondrial membrane potential.
5 – Cellular protrusions as wave propagations coupled with membrane curvatures.
Yiyan Lin, Siyu Ye, Saki Takayanagi, Takanari Inoue, Huaqing Cai, Miho Iijima, Mike Piacentino, Peter N. Devreotes
Departments of Cell Biology, Johns Hopkins School of Medicine, Baltimore, MD
Institute of Biophysics, Chinese Academy of Sciences, Beijing, China
Cell migration is a vital process in embryonic development and morphogenesis. While cells typically move by extending actin-based protrusions at the front and contracting at the rear, they can form a wide variety of protrusions—such as lamellipodia, filopodia, pseudopodia, macropinosomes, and blebs. This diversity has often been attributed to distinct molecular pathways. However, emerging evidence points to a common underlying mechanism involving cortical signaling waves—spatiotemporal propagations of signaling activity across the cell cortex—increasingly linked to protrusion dynamics. Recently, we showed that small, reversible perturbations to the signaling network can rapidly switch cells between cup-like and lamellipodia-like protrusions, corresponding to predictable changes in the size and duration of cortical waves. These findings suggest that different protrusions may exist along a continuum, governed by a shared excitable signaling system. In this study, we further demonstrate that cortical waves are intimately coupled to dynamic membrane fluctuations, challenging the longstanding view that such waves are confined to flat cell surfaces. We identify key proteins—F-BAR domain proteins and myosin I motors—that help generate and regulate these waves. Our findings support a unified model in which signaling waves coordinate the formation of diverse protrusions, offering new insights into how cells modulate their shape and movement.
6 – Sst2 is essential for pre-morphogenic gradient sensing in mating yeast
Alanda Kelly; David Stone
University of Illinois, Chicago
Chemotropism is directed cell growth in response to chemical gradients. Although this process has been studied for decades, how cells accurately interpret and respond to shallow, complex, and dynamic chemical gradients is not fully understood. The yeast mating response is chemotropic: cells of opposite mating type signal their position to potential partners by secreting mating pheromones. We have proposed a deterministic gradient sensing model that explains how yeast cells detect and orient toward their mating partners. Using an intrinsic polarity site, cells assemble a gradient-tracking machine (GTM) at the plasma membrane composed of signaling, polarity, and trafficking proteins. A key function of this system is the activation of G-protein coupled receptors and their associated G-proteins. Activation of this complex enables the GTM to direct vesicle delivery and carry new receptor and G-protein toward the gradient source in a process called tracking. The negative regulator of G-protein signaling, Sst2, catalyzes the inactivation of the Gα subunit. Our results indicate that Sst2 plays a critical role in gradient tracking. We have postulated two distinct ways that Sst2 contributes to GTM function, but these are not easily distinguished by experiment. Hence, a computational model of yeast gradient tracking is of immediate importance.
7 – Engineering a genomically recoded organism with one stop codon
Michael W. Grome1,2, Michael T. A. Nguyen1,2, Daniel W Moonan2,3, Kyle Mohler2,4, Kebron Gurara1,2, Shenqi Wang1,2, Colin Hemez1,2,5, Benjamin Stenton1,2, Yunteng Cao1,2, Felix Radford1,2, Maya Kornaj1,2,4, Jaymin Patel1,2, Maisha Prome1,2, Svetlana Rogulina2,4, David Sozanski1,2, Jesse Tordoff1,2, Jesse Rinehart2,4, Farren J. Isaacs1,2,5
1. Department of Molecular, Cellular & Developmental Biology, Yale University, New Haven, CT 06520, USA 2. Systems Biology Institute, Yale University, West Haven, CT 06516, USA
3. Department of Molecular Biophysics & Biochemistry, Yale University, New Haven, CT 06520, USA
4. Department of Cellular & Molecular Physiology, Yale University, New Haven, CT 06520, USA
5. Department of Biomedical Engineering, Yale University, New Haven, CT 06520, USA
The genetic code is conserved across all domains of life, yet exceptions have revealed variations in codon assignments and associated translation factors 1-3 . Inspired by this natural malleability, synthetic approaches have demonstrated whole-genome replacement of synonymous codons to construct genomically recoded organisms (GROs) 4,5 with alternative codes. However, no efforts have fully leveraged translation factor plasticity and codon degeneracy to compress translation function to a single codon and assess the possibility of a non-degenerate code. Here, we describe construction and characterization of the first GRO - “Ochre” – to fully compress a translational function into a single codon. We replaced 1,195 TGA stop codons with synonymous TAA within ∆TAG Escherichia coli C321.∆A 4 . We then engineered release factor 2 and tRNA Trp to mitigate native UGA recognition, translationally isolating four codons for non-degenerate functions. This rendered UAA as the sole stop codon, UGG for Tryptophan, and reassignment of UAG and UGA for multi-site incorporation of two distinct non-standard amino acids into single proteins with >99% accuracy. Ochre fully compresses degenerate stop codons into a single codon, presenting an important step toward a 64-codon non-degenerate code while enabling precise production of multi-functional synthetic proteins possessing unnatural encoded chemistries with broad utility in biotechnology and biotherapeutics.
8 ‐ Generative Vision-Based Modeling for Mechanistic Inference on Spatial Dynamical Data
Jun Won Park, Kangyu Zhao, and Sanket Rane
Irving Institute for Cancer Dynamics, Columbia University, New York, NY, USA
Biological systems commonly exhibit complex spatiotemporal patterns whose underlying generative mechanisms pose a significant analytical challenge. Traditional approaches to spatiodynamic inference rely on dimensionality reduction through summary statistics, which sacrifice complexity and interdependent structure intrinsic to these data in favor of parameter identifiability. This imposes a fundamental constraint on reliably extracting mechanistic insights from spatiotemporal data, highlighting the need for analytical frameworks that preserve the full richness of these dynamical systems. To address this, we developed a simulation-based inference framework that employs vision transformer-driven variational encoding to generate compact representations of the data, exploiting the inherent contextual dependencies. These representations are subsequently integrated into a likelihood-free Bayesian approach for parameter inference. The central idea is to construct a fine-grained, structured mesh of latent representations from simulated dynamics through systematic exploration of the parameter space. This encoded mesh of latent embeddings then serves as a reference map for retrieving parameter values that correspond to observed data. Through the integration of generative modeling and Bayesian principles, our approach provides a unified inference framework that captures and analyzes both spatial and temporal patterns that manifest in multivariate dynamical systems.
9 - Modeling reveals the strength of weak interactions in stacked trimeric ring assembly
Leonila Lagunes, Eric J. Deeds
University of California, Los Angeles
Cell and organismal viability rely on macromolecular machines regulating many vital processes. Interestingly, cells cannot synthesize these macromolecular machines as functioning units. Instead, they synthesize the molecular parts that must then assemble into the functional complex. An extremely common motif is a stacked ring-like topology, such as in the proteasome or the ubiquitin conjugating enzyme E2. Thus, understanding how stacked trimers assemble is crucial for our understanding of how complexes are regulated. In this work, we developed a mathematical model of stacked trimer assembly that accounts for different binding affinities between and within rings. Our main finding is that deadlock – a severe form of kinetic trapping– can be extremely long, lasting for days or longer. Deadlock is worst when all the interfaces have high binding affinities. Therefore, we predict that evolutionary pressures select against stacked trimers having strong binding affinities throughout. We tested our prediction by analyzing solved stacked trimer structures; we found that indeed the majority – if not all – of the stacked trimers did not contain very strong interactions throughout. Finally, to better understand the origins of deadlock, our pathway analysis shows that when all the binding affinities are strong, many of the possible pathways are utilized, consuming subunits, and creating high levels of deadlock. In sum, our work provides critical insight into the evolutionary pressures that have shaped the assembly of stacked rings.
10 – MEMBRANE SEGMENTATION IN NOISY DATA VIA PHYSICS INFORMED NEURAL NETWORKS
Atsushi Matsuda1, Christopher T. Lee2, Matthew Akamatsu1
1.University of Washington
2.University of California, San Diego
Membrane segmentation is a critical step in the analysis of cryo-electron tomography data, enabling structural and biophysical insights into cellular organelles. While recent advances in image processing and deep learning have enabled automated segmentation, these methods often struggle under low signal-to-noise conditions. Enhancing segmentation robustness in such environments is therefore essential for accurate downstream analysis. We hypothesized that incorporating physical principles into the segmentation process could improve output quality. Biological membranes do not form arbitrary structures—their shapes and deformations are governed by well-established physical laws of membrane mechanics. We reasoned that integrating these constraints into the segmentation task could guide the model toward more accurate membrane identification in noisy data. To test this idea, we developed a physics informed neural network (PINN) that reconstructs membrane structures from tomographic images. Rather than relying solely on image features, the model is trained to satisfy both image fidelity and physical consistency, where the latter is defined by the principles of membrane mechanics. We evaluated the model’s performance across varying levels of image noise and found that incorporating physical constraints significantly improves segmentation quality under noisy conditions. This finding suggests a new approach to enhance the robustness of membrane segmentation, which could be incorporated into applied cryo-ET analysis pipelines.
11 – Integrating qualitative data via mathematical modeling reconciles discordant observations and offers a candidate mechanism for intracellular regulation of BRUTUS in Arabidopsis
Ghizelle Jane E. Abarro, Dipali Srivastava, DT Flaherty, Terri A. Long, and Belinda S. Akpa
University of Tennessee Knoxville
Iron (Fe) is an essential nutrient but becomes toxic when present in excess. To regulate Fe levels, living systems depend on internal sensing mechanisms. In Arabidopsis thaliana, the putative Fe sensor BRUTUS (BTS) directly binds Fe, and BTS knockdown mutants accumulate more Fe – suggesting that the protein negatively regulates Fe uptake. Given this evidence, one might expect down-regulation of BTS expression under Fe-deficient conditions. Instead, BTS is upregulated under Fe deficiency. Experimental efforts to elucidate the interplay between Fe status and BTS activity have yielded discordant observations and prompted contradictory conclusions. In vitro studies have reported BTS destabilization as Fe levels increase. By contrast, live-cell imaging reveals increased cytosolic BTS as Fe levels increase. Furthermore, mutations that partially abrogate Fe binding produce divergent levels of cytosolic and nuclear BTS – with pronounced decreases observed in the cytosol and modest increases observed in the nucleus. Using mathematical modeling as a formal, testable sense-making strategy, we sought to determine whether there exists a kinetic regime under which Fe-responsive stability and translocation of BTS could concurrently explain this group of empirical observations. By embedding candidate subcellular events into a system of ordinary differential equations, we modeled BTS accumulation in response to Fe availability. Using simulation-based inference to match model predictions to qualitative, in vivo observations, we identified broad ranges of plausible kinetics that pointed to Fe-binding stoichiometry as a key determinant of BTS fate. We pursued this hypothesis – and others derived from simulations spanning the domain of plausible kinetics – by performing a series of model-driven experiments. Results from these experiments allowed us to further constrain the credible parameter space such that an initial pool of 120k plausible values was reduced to just 34. That is, less than 0.03% of the parameter space identified via inference could account for the collective emergent features of BTS behavior in vivo and in vitro. Through this iterative modeling-experimentation loop, we reconciled seemingly contradictory empirical observations via a novel systems mechanism wherein BTS persistence varies with Fe-binding stoichiometry. Supported by further validation of our model’s predictions, we posit that BTS abundance is a non-monotonic function of Fe status, defined by competing effects of Fe-mediated stability, proteasomal degradation, and Fe-stimulated nuclear translocation.
12 – Mechanisms of Protein Self-Assembly: From Microtubule Dynamics to Membrane Localization
Smriti Chhibber, Margaret Johnson
Department of Biophysics, Johns Hopkins University
Microtubules are dynamic cytoskeletal filaments whose growth and shrinkage are tightly regulated by GTP hydrolysis. We simulate the growth of individual protofilaments composed of two distinct tubulin dimers on a 2D lattice using a Gillespie algorithm that accounts for lateral and longitudinal interactions, as well as hydrolysis dynamics. This model allows us to examine how variations in binding energies and hydrolysis rates influence filament stability, catastrophe frequency, and rescue events. In parallel, we investigate how biological assemblies are enhanced by membrane localization, where confinement to a 2D surface increases effective molecular concentration and promotes self-assembly. Using a thermodynamic framework, we compare protein assembly in 3D solution versus 2D membrane environments, revealing how confinement and surface affinity drive equilibrium toward membrane-bound states. To complement our theoretical models, we use NERDSS, a spatial rule-based simulation platform, to capture multivalent self-assembly in both 3D and 2D contexts. NERDSS enables us to incorporate spatial geometry and binding rules, providing dynamic insight into how protein complexes form and organize across dimensional boundaries.
13 – Functional implications of biomolecular condensate size distribution
Aniruddha Chattaraj, Eugene I. Shakhnovich
Department of Chemistry and Chemical Biology, Harvard University, Cambridge 02138, MA, USA
Biomolecular condensates are phase separated sub-cellular structures that spatially control the biochemistry for a wide array of systems. For a given condensate type, the existence of multiple droplets inside living cells is a peculiar phenomenon not compatible with the predictions of equilibrium statistical mechanics. Thermodynamics of polymer phase separation predicts that, beyond a threshold concentration, the system would demix into a polymer-dense phase (single large condensate), surrounded by a solvent-rich dilute phase. However, with in-vitro and cellular experiments, we always observe multiple condensates that persist over the experimental timespan. In this work, we attempt to address the problem of multiple condensates state (MCS) from a functional perspective. We combined Langevin dynamics, reaction-diffusion simulation, and dynamical systems theory to show that MCS can indeed be a functional strategy to optimize the desired biochemical output. Using Arp2/3 mediated actin nucleation pathway as an example, we showed that actin polymerization is maximum at an optimal number of condensates. The peripheral location of Arp2/3 and differential diffusion of actin in monomeric and polymeric form makes MCS functionally more beneficial. For a fixed amount of Arp2/3, MCS produces a greater response compared to its single condensate counterpart. Density of Arp2/3 at the condensate surface serves as a bifurcation parameter to create a functional switch that can be reversibly tuned. Our analysis reveals the functional significance of the condensate size distribution which can be mapped to the recent experimental findings. Since non-linearity is a ubiquitous feature of intracellular molecular networks, we envision that MCS could serve as a generic functional strategy and structures of network motifs may have evolved to accommodate such configurations.
14 – Enzyme Structure and Function at Extreme Temperatures
Catherine Le (presenter), Maanasa Panuganti, and Melanie Cocco
Dept Molecular Biology and Biochemistry, UC Irvine
Bacterial enzymes have evolved to function across a temperature range exceeding 110 °C, offering valuable insights to industrial enzyme engineering, biotherapeutic stabilization, genomics, and protein structure prediction. Remarkably, these enzymes often retain structural similarity while adapting their amino acid sequences to optimize stability and activity at their native environmental temperatures. For example, a DNA polymerase from Arctic bacteria has high accuracy of base incorporation at –19 °C but loses fidelity at room temperature, whereas thermophilic polymerases can lose catalytic ability at low temperatures even though they have the same structural fold as enzymes evolved for low temperatures. Since DNA polymerases are among the most ancient enzymes, studying their adaptation provides a unique window into how amino acid selection drives stability and function under extreme conditions. Our focus is Pol IV, a bypass polymerase expressed across prokaryotes from polar ice to deep-sea thermal vents. Pol IV contains four independent domains. Using NMR spectroscopy, we have studied Pol IV from a thermophile (optimal at 75–90 °C) and an enteric bacterium (optimal at 37 °C). The thermophilic enzyme remains structured and active at high temperatures but cold-denatures near room temperature. NMR dynamics measurements revealed a range of domain-specific stability: Palm > LF > Fingers > Thumb. We have compiled over 2,700 Pol IV sequences from bacteria thriving within four major temperature ranges, spanning more than 100 °C. Our future goal is to use AI tools to identify sequence patterns conserved within each temperature range and, most significantly, differences between temperatures. These results will guide domain swapping and the directed evolution of Pol IV variants for extreme hot or cold conditions. Resulting variants will be tested for structure, function, and stability. Since enzymes must balance structural integrity with conformational flexibility required for function, we will assess structure, dynamics, thermal denaturation, and catalytic rates/fidelity.
15 – The effect of a pheromone protease on yeast gradient sensing
Paul A. Urban and David E. Stone
Department of Biological Sciences, University of Illinois at Chicago
The mating response of the budding yeast S. cerevisiae relies on chemotropism, a cellular process relevant to many aspects of biology. Haploid yeast cells of opposite mating types detect pheromone gradients produced by one another and polarize their growth towards the gradient source. For accurate gradient sensing, yeast cells assemble a gradient tracking machine (GTM) composed of sensory proteins and secretion machinery, which incrementally redistributes upgradient along the cell membrane until it faces the point of maximal pheromone concentration in a process called tracking. While extensive studies have documented how signaling mechanisms amplify the intracellular signaling gradient to enable tracking, the effects of external pheromone gradient steepening have not yet been explored. We are investigating how Bar1, a protease that degrades the pheromone α-factor, affects the yeast cell’s ability to gradient sense. We found that Bar1 is essential for gradient sensing and suggest that Bar1 enables wild-type levels of tracking through a cell-autonomous mechanism. These results support the hypothesis that Bar1 steepens the pheromone gradient along the GTM, enabling accurate pre-morphogenic gradient sensing.
16 – A variational deep-learning approach to modeling memory T cell dynamics
Christiaan H. van Dorp1, Joshua I. Gray2, Daniel H. Paik2, Donna L. Farber2, Andrew J. Yates1
1.Department of Pathology and Cell Biology, Columbia University Irving Medical Center, New York City, NY, USA
2.Department of Microbiology and Immunology, Columbia University Irving Medical Center, New York City, NY, USA
Mechanistic models of dynamic, interacting cell populations have yielded many insights into the growth and resolution of immune responses. Historically these models have described the behavior of pre-defined cell types based on small numbers of phenotypic markers. The ubiquity of deep phenotyping therefore presents a new challenge; how do we confront tractable and interpretable mathematical models with high-dimensional data? To tackle this problem, we studied the development and persistence of lung-resident memory CD4 and CD8 T cells (TRM) in mice infected with influenza virus. We developed an approach in which dynamical model parameters and the population structure are inferred simultaneously. This method uses deep learning and stochastic variational inference and is trained on the single-cell flow-cytometry data directly, rather than on the kinetics of pre-identified clusters. We show that during the resolution phase of the immune response, memory CD4 and CD8 T cells within the lung are phenotypically diverse, with subsets exhibiting highly distinct and time-dependent dynamics. TRM heterogeneity is maintained long-term by ongoing differentiation of relatively persistent Bcl-2hi CD4 and CD8 TRM subsets which resolve into distinct functional populations. Our approach yields new insights into the dynamics of tissue-localized immune memory, and is a novel basis for interpreting time series of high-dimensional data, broadly applicable to diverse biological systems.
17 – Defining and programming transcriptional activatability in bacterial promoters
Debora Tenenbaum, Chirangini Pukhrambam, Andalus Ayaz, Bryce Nickels, Justin B. Kinney
Simons Center for Quantitative Biology, Cold Spring Harbor Laboratory, NY, USA Waksman Institute of Microbiology, Rutgers University, Piscataway, NJ, USA.
Transcription initiation is the primary control point of gene expression in bacteria, yet the sequence-to-function relationship governing this process remains incompletely understood. We investigate how promoter DNA sequences encode regulatory information by systematically characterizing transcription initiation by E. coli RNA polymerase (RNAP) using in vitro transcription assays on diverse promoter libraries. By sequencing the 5’ ends of transcripts, we quantify initiation rates at base-pair resolution across multiple RNAP concentrations. By varying RNAP concentrations in our in vitro transcription assays, we can dissect how specific promoter sequence elements modulate distinct mechanistic steps of transcription initiation. To interpret the resulting data, we plan to combine biophysically motivated models with flexible deep learning approaches—aiming to facilitate mechanistic interpretation of sequence effects while capturing complex dependencies beyond current mechanistic understanding.
18 – Multiple approaches to identifying tissue correlates of serum antibody/neutralization titers against SARS-CoV-2
Juliane Schröter1, Christiaan H van Dorp1, Julia M Davis-Porada2, Donna L Farber2, Andrew J Yates1
1 Columbia University Irving Medical Center, Department of Pathology and Cell Biology, New York, NY,
2 Columbia University Irving Medical Center, Department of Microbiology and Immunology, New York, NY
Background: Infections and vaccinations elict both humoral and cellular immune responses. While systemic antibody titers are widely used as correlates of protection, the contribution of tissue-resident memory T and B cells to protective immunity remains unclear. We investigated how SARS-CoV-2-specific immune responses across tissues relate to systemic binding and neutralizing antibody levels in vaccinated individuals. Methods: We analyzed samples from 57 deceased SARS-CoV-2-vaccinated organ donors – 20 uninfected and 37 previously infected. Spike-specific T and B cells were profiled across seven tissues. Serum IgG binding to Spike and receptor-binding domain (RBD), along with neutralizing titers against ancestral WA1, Delta, and Omicron BA.2.12.1 variants, were measured. Due to incomplete tissue sampling and heterogeneous cellular coverage, we performed multivariate imputation to generate 100 complete datasets. Consensus findings were derived using a range of analytical approaches, including correlation analysis, principal component analysis , canonical correlation analysis, multiple regression, random forest, and structural equation modeling. Results: Antibody titers varied widely and were strongly associated with tissue-resident immune features, notably memory B cells and CD8⁺ T cells in lung-associated lymph nodes (LLN), lung, and spleen. Surprisingly, higher antibody titers were inversely associated with splenic T follicular helper cells. Key predictors of humoral immunity included infection history, vaccine dose count, and time into the pandemic, especially for variant-specific responses to Delta and Omicron. Data imputation reduced the predictive power of tissue-derived variables, underscoring the impact of sampling limitations on interpretability. Conclusion: These findings highlight the individualized nature of SARS-CoV-2 vaccine responses and the added value of tissue-based immune profiling. Furthermore, they emphasize the importance of integrating computational approaches with tissue analyses to identify robust correlates of long-term immunity and to uncover mechanisms underlying vaccine efficacy – suggesting that T-cell immunity may need to be assessed independently from humoral responses.
19 – Blood Digital Twins Developed Using Dynamic State Modeling of Single-Cell RNA-seq Data
Pancy Lwin, Juilee Thakar
URMC Department of Microbiology and Immunology, URMC Department of Biostatistics & Computational Biology, URMC Department of Biomedical Genetics
The development of blood digital twins represents a critical step toward personalized medicine and in-silico clinical trials. Here, we present a dynamic modeling framework that leverages single-cell RNA-seq data to construct immune digital twins through Boolean network-based State Transition Graphs (STGs). By modeling intracellular signaling dynamics, we identify dominant attractors that represent stable cellular states and track pseudotime progression across disease-specific trajectories. Cells are clustered based on attractor similarity, spatial proximity in reduced dimensions, and alignment with clinical phenotypes (e.g., AS+ vs AS−). Integrating dynamic signaling models with pseudotime inference enables estimation of transition probabilities and identification of key driver genes shaping disease fate. In silico perturbation experiments reveal gene targets that significantly shift disease trajectories, offering insights into therapeutic vulnerabilities. This personalized modeling approach builds upon the scBONITA (single-cell Boolean Omics Network Invariant-Time Analysis) framework from Palshikar et al., to generate patient-specific profiles that summarize attractor landscapes, driver genes, and immune dynamics. Looking forward, expanding this framework to encompass broader system level processes will bring us closer to implementing virtual clinical trials, transforming the future of drug development and precision immunotherapy. Citation: Palshikar, M.G., Palli, R., Tyrell, A. et al. Executable models of immune signaling pathways in HIV-associated atherosclerosis. npj Syst Biol Appl 8, 35 (2022). https://doi.org/10.1038/s41540-022-00246-5
20 – A Genome-Complete Foundation For A Whole Human Epithelial Cell Model
Jonah R. Huggins1,2, Atalanta Harley-Gasaway1, Marc R. Birtwistle1
1. Department of Chemical and Biomolecular Engineering, Clemson University
2. School of Computing, Clemson University
Whole cell models have been described for multiple single-celled organisms but not yet for a human cell. We previously reported one of the largest models of a human epithelial cell that captures key proliferation and death pathways (~150 genes) and single-cell heterogeneity using custom simulation algorithms. In a key step towards a whole human epithelial cell model, we are creating species for twenty-thousand functional human genes and their nascent products (mRNA, proteins), resulting in a genome-complete foundation. Three primary considerations arise due to this model size: data organization, runtime, and a means to evaluate its accuracy. For the first, we have integrated a standard species naming convention to mitigate conflicting unique identifiers and generate separate stochastic and deterministic SBML files for clean and scalable simulation. For the second, previous work highlighted that our dual Python/C++ simulation framework creates long runtime due to communication bottlenecks. To speed simulations, the solvers were rewritten in C++. For the third, larger models require ever increasing amounts of data, but comparison between the two remains non-systematic. We created a systematic benchmarking tool using PEtab for data organization, and topologic, multiprocessing algorithms to organize simulations based on the required conditions (e.g. drugs, growth factors, times, number of single cells). This work provides a foundation for a whole human epithelial cell model whereby not only our lab but also the modeling community could potentially contribute to progress.
21 – Synergistic bactericidal pore formation by differential targeting membranes by histones and AMPs
Yonghan Wu
Department of Physics and Astronomy, University of California, Irvine
Histone and antimicrobial peptides (AMPs) are crucial components of innate immunity, contributing to host defense by neutrophil extracellular traps and lipid droplets. But the mechanism of its antimicrobial functions remains poorly understood. Using stochastic optical reconstruction microscopy (STORM), we visualize the localization of histone and AMP at single-molecule resolution. Our cryo-electron microscopy (cryo-EM) shows that histone or AMPs alone do not damage membrane but together they form bactericidal pores, which we define as synergy. To elucidate the kinetics, we developed a mathematical model of molecules translocation and pore formation across the four leaflets of the bacterial inner and outer membranes. We found that the synergy arises from differential targeting of the membrane leaflets, where histones targeting the inner membrane leaflet and AMPs the outer leaflets. In summary, we report synergistic antimicrobial behavior between histone and AMPs against E. coli by differential targeting, providing insights into potential therapeutic strategies against gram-negative bacteria.
22 – Metapages: A Platform for Reproducible, Shareable, Interactive, AI-Enhanced Mechanistic Modeling in the Browser
Dion Whitehead
metapages, LLC
As AI systems begin to rival domain experts in predictive performance, mechanistic modeling is under pressure—yet remains indispensable. We present the *metapage platform*, a web tool that helps modelers, experimentalists, and AI developers co-create executable, visual scientific simulations entirely in the browser. The metapage platform addresses several pain points in mechanistic modeling: reproducibility, accessibility, and collaboration. The metapage platform lets users link datasets, code, and containerized compute environments to visualize dynamic biological models, AI predictions, and hybrid workflows in a modular way. AI models can be plugged in not as black boxes, but as swappable components with interpretable inputs and outputs. This fosters a new kind of transparent “middle-out” modeling, where human insight and machine learning reinforce one another. The platform supports live computation (e.g., simulation, optimization, structure prediction), version-controlled model sharing, and multimodal visualization—without requiring any local installation. This makes it easy for modelers to collaborate with experimentalists, iterate quickly, and bring AI capabilities into traditional modeling loops, such as hypothesis testing, parameter estimation, and model selection. The platform is built on an open-source framework, and is extensible. Early use cases include protein design, cellular signaling, and morphogenesis modeling—each blending mechanistic insight with AI acceleration. We argue that the future is not a binary choice between mechanistic modeling and AI, but a convergence. The metapages platform empower modelers to remain central—designing, debugging, and validating models—while leveraging the scale and generative power of AI, and sharing fully executable models with collaborators and the wider public.
23 – A Computational and Experimental Investigation of Cell-Cell Interactions Driving Tumor-Induced Bone Disease
Alexandra Gutierrez Vega1*, Natalie E. Bennett2*, Saja Alshafeay1, Erik P. Beadle3, Julie A. Rhoades2,3, and Leonard A. Harris1,4,5
1 Interdisciplinary Graduate Program in Cell & Molecular Biology, University of Arkansas, Fayetteville, AR
2 Program in Cancer Biology, Vanderbilt University, Nashville, TN
3 Division of Clinical Pharmacology, Department of Medicine, Vanderbilt University Medical Center, Nashville, TN
4 Department of Biomedical Engineering, University of Arkansas, Fayetteville, AR
5 Cancer Biology Program, Winthrop P. Rockefeller Cancer Institute, University of Arkansas for Medical Sciences, Little Rock, AR
*Equal contributors
Tumor-induced bone disease (TIBD) is a complex and poorly understood condition that arises when tumors metastasize to the bone. The prevailing model of TIBD is known as the “vicious cycle” model, in which tumor cells produce factors, like parathyroid hormone-related protein (PTHrP), that promote bone destruction by osteoclasts (OCs). This releases growth factors, e.g., TGF-beta, which enhance tumor growth and further exacerbate bone damage. Current treatments, including bisphosphonates and RANKL inhibitors, reduce bone loss but fail to halt tumor progression or improve survival. Hence, there is a critical need to develop a deeper understanding of the biological mechanisms driving TIBD to develop more effective therapies. Here, we present a computational model of osteoblast (OB) and OC interactions with tumor cells in the bone microenvironment. The model includes logistic tumor growth, TGF-beta enhancement of tumor proliferation and PTHrP synthesis, and OB inhibition and OC activation by tumor-secreted factors. Parameter values were estimated by fitting the model to in vivo time-course data for bone density and OB, OC, and tumor cell counts from untreated and drug-treated mice injected with parental and bone-metastatic tumor cells. For untreated parental tumors, the model predicts reduced OC growth relative to untreated bone-metastatic tumors. In response to the OC inhibitor zoledronic acid (ZA), the model predicts a range of outcomes for bone-metastatic tumors, from no response to significant growth inhibition. Overall, our results provide insights into the basic mechanisms underlying TIBD and raise questions regarding a fundamental assumption of the vicious cycle model, i.e., that tumor cells depend on bone-derived growth factors for survival. Future work will extend this approach to other tumor types and drugs and apply dynamic pathway analysis to identify novel intervention points for maximizing tumor inhibition while preventing bone destruction.
Page updated
Report abuse