Interdisciplinary Center for Quantitative Modeling in Biology

PLENARY SPEAKERS

 

Dec. 11th

Mettine Bos
Division of Thrombosis and Hemostasis, Leiden University Medical Center, Director, Einthoven Laboratory for Vascular and Regenerative Medicine,The Netherlands, Chair, European Thrombosis & Hemostasis Alliance

Selective Tuning of Blood Coagulation Proteases to Treat Bleeding

Life-threatening bleeding events are substantial complications for patients with a compromised blood coagulation response. The inability to clot efficiently can result from treatment with anticoagulants or is caused by a congenital or acquired bleeding disorder. Dr. Bos aims to engineer innovative variants of blood coagulation proteases to improve treatment of bleeding. Her research explores evolutionary adaptations in snake venom coagulation factors to identify modifications that could enhance the functionality of blood coagulation proteases. This has led to the development of factor X variants that maintain procoagulant activity but are resistant to direct Xa inhibitors, creating potential reversal agents for anticoagulant drugs in emergencies, such as life-threatening bleeding. One such variant is advancing in clinical development. Another area of her research involves engineering factor IX variants that operate independently of factor VIII, presenting a promising treatment approach for hemophilia A. To support these efforts, Dr. Bos uses cutting-edge computational techniques, including AlphaFold and molecular dynamics simulations, for protein structure and interaction modeling. These tools, combined with biochemical characterization and preclinical validation, enable the precise tuning of serine proteases to produce prohemostatic agents, thereby addressing unmet needs in the treatment of bleeding.
 

Dec. 12th

Holly Goodson
Department of Chemistry and Biochemistry, University of Notre Dame, Fellow of the American Society for Cell Biology

Developing a Multiscale Understanding of Energy-utilizing Polymers

Microtubules are energy-utilizing polymers that serve as essential components of the eukaryotic cytoskeleton, enabling cells to divide, segregate their chromosomes, and organize themselves internally. Microtubule filaments display a puzzling behavior known as dynamic instability, in which they undergo distinct phases of growing and shortening, with approximately random transitions in between. Our lab and its collaborators integrate experiments, simulations and mathematical modeling to build a multi-scale and predictive understanding of microtubule dynamics.  In this seminar, I will discuss our efforts to use machine-learning to objectively quantify and characterize MT dynamics, our application of this software to dissect the mechanism of a MT binding protein, and some “big picture” biological conclusions drawn from analysis of emergent population-level behaviors exhibited by our dimer-scale, agent-based simulations of microtubule assembly.

INVITED SPEAKERS
 

Maziar Raissi, Assistant Professor, Department  of Mathematics, UCR

Data-Efficient Deep Learning Using Physics-Informed Neural Networks

This work addresses the challenge of integrating conservation laws, physical principles, and phenomenological behaviors—expressed by differential equations—with the vast datasets available in engineering, science, and technology. At the intersection of probabilistic machine learning, deep learning, and scientific computation, the aim is to create a framework that combines classical methods in applied mathematics and mathematical physics with modern data-driven approaches. The research explores two main directions: 1. Developing data-efficient learning models that leverage physical laws, represented by time-dependent and nonlinear differential equations, to identify patterns in high-dimensional experimental data. 2. Designing numerical algorithms that blend equations with noisy, multi-fidelity data, infer latent quantities of interest (e.g., solutions to differential equations), and quantify computational uncertainty.

Siting Liu, Assistant Professor, Department  of Mathematics, UCR

In-Context Learning for Differential Equations

This talk presents In-Context Operator Networks (ICON), a novel neural-network-based framework designed to learn and apply operators directly from prompted data during inference, without requiring any weight updates. Traditional methods rely on neural networks to approximate solutions for specific equations or operators, necessitating retraining or fine-tuning for new problems. In contrast, ICON trains a single neural network to serve as a general operator learner, enabling it to adapt to new problems with minimal effort. By leveraging shared structures across operators, ICON requires only a few demonstration examples in the prompt to learn a new operator effectively. Our results demonstrate ICON's ability as a few-shot operator learner across a diverse range of differential equation problems, including forward and inverse tasks for ordinary differential equations, partial differential equations, and mean-field control problems. Furthermore, we highlight ICON's generalization capabilities, showcasing its potential as a powerful tool for solving complex operator learning challenges in scientific computing and beyond. This is a joint work with Liu Yang (NUS), Tingwei Meng (UCLA), and Stanley Osher (UCLA).
Yiwei Wang, Assistant Professor, Department  of Mathematics, UCR

Energetic Variational Neural Network Discretizations of Gradient Flows

Numerous applications in physics, material science, biology, and machine learning can be modeled as gradient flows. In this talk, we present a structure-preserving Eulerian algorithm for solving L2-gradient flows and a structure-preserving Lagrangian algorithm for solving generalized diffusions by employing neural networks as tools for spatial discretization. Unlike most existing methods that construct numerical discretizations based on the strong or weak form of the underlying PDE, the proposed schemes are constructed based on the energy-dissipation law directly. This guarantees the monotonic decay of the system's energy, which avoids unphysical states of solutions and is crucial for the long-term stability of numerical computations. To address challenges arising from nonlinear neural-network discretization, we adopt a temporal-then-spatial discretization approach on these variational systems. The proposed neural-network-based schemes are mesh-free, allowing us to solve gradient flows in high dimensions in biology and machine learning.
Vivian Li, Assistant Professor, Department of statistics, UCR

Detection and interpretation of spatial gene expression variation

Recent advances in spatially resolved transcriptomics technologies have opened up new avenues for understanding gene expression heterogeneity in spatial contexts, and an important task in spatial transcriptomics analysis is the identification of spatially variable genes (SVGs). While various computational methods exist for SVG detection, most focus solely on statistical significance and have limitations in capturing continuous expression patterns across spatial domains and incorporating cell/spot-level covariates. To address these challenges, we introduce spVC, a novel statistical method to detect and interpret SVGs based on a generalized Poisson model. spVC integrates constant and spatially varying effects of cell/spot-level covariates, enabling comprehensive exploration of gene expression variability and enhancing interpretability. It provides a convenient tool to identify potential factors that contribute to gene expression variability, including spatial locations and other cell/spot- level covariates such as cell types or tissue layers. In summary, spVC is a versatile tool for the identification, interpretation, and comprehension of gene expression variation in spatial transcriptomics data.
Yuzhou Chen, Assistant Professor, Department of Statistics, UCR

Topological Deep Learning in Computer-Aided Drug Discovery and Beyond

In computer-aided drug discovery (CADD), virtual screening (VS) is used for identifying the drug candidates that are most likely to bind to a molecular target in a large library of compounds. Most VS methods to date have focused on using canonical compound representations or generating alternative fingerprints of the compounds by training progressively more complex variational autoencoders (VAEs) and graph neural networks (GNNs). Although VAEs and GNNs led to significant improvements in VS performance, these methods suffer from reduced performance when scaling to large virtual compound datasets. The performance of these methods has shown only incremental improvements in the past few years. To address this problem, we developed a novel method using multiparameter persistence (MP) homology that produces topological fingerprints of the compounds as multidimensional vectors. We further establish theoretical guarantees for the stability properties of our proposed MP signatures, and demonstrate that our models, enhanced by the MP signatures, outperform state-of-the-art methods on benchmark datasets by a wide and highly statistically significant margin.
Vishwanath Saragadam, Assistant Professor, Department of Electrical and Computer Engineering, UCR

Signal Processing-Inspired AI

We are now in a data-driven era where large datasets are used to train AI models to solve challenging problems, such as 2D and 3D computed tomography, hyperspectral imaging, image denoising, and so on. However, what do we do when we do not have access to such large datasets? Classical data-free approaches to solving such problems involved leveraging the structure of the data -- this included sparsity of gradients in images, wavelets to model strong edges, low rank structure of hyperspectral images, and so on. These models however do not faithfully capture all the intricate details. In this talk, I will discuss how classical signal processing techniques can be used to supercharge AI models to solve problems in data scarce regimes. First, I will talk about representing low-rank data as outputs of an untrained deep network, which enables strong regularization as a result of the structure of the deep network. Second, I will talk about how a complex wavelet can be used as an activation function in certain deep network-based problems to enable strong regularization, and high representation capacity. To this end, we will see that the classical insights into data representations are still relevant, and can be used to overcome some of the crucial challenges associated with data-driven AI-based approaches.

Wei-Chun Chou, Assistant Professor, Department of Environmental Sciences, UCR

Advancing Nanomedicine Through Artificial Intelligence (AI)-Driven Computational Modeling and Applications

The integration of artificial intelligence (AI) and machine learning methodologies into computational modeling frameworks has revolutionized the evaluation and prediction of pharmacokinetics, providing transformative insights not only for small-molecule drugs but also for the complex landscape of nanomedicine. This presentation highlights the pivotal role of AI-assisted approaches in advancing the prediction of nanomedicine’s absorption, distribution, metabolism, and excretion (ADME) properties, addressing critical challenges posed by their unique physicochemical characteristics and interactions with biological systems. By coupling AI with physiologically based pharmacokinetic (PBPK) modeling, we introduce a novel paradigm that enhances drug delivery strategies’ precision, scalability, and efficiency. This synergy enables the identification of optimal formulations, tailoring of therapeutic interventions, and reduction of experimental burden, ultimately fostering the development of next-generation nanomedicines with improved therapeutic outcomes. The discussion also explores future directions, including the application of multi-modal data integration, neural network-driven simulations, and AI-enhanced predictive modeling to drive innovation in precision nanomedicine.
Shahab Vahdat, Assistant Professor, Department of Bioengineering, UCR

Integrating Brain and Spinal Cord fMRI: New Frontiers in Functional Neuroimaging and Connectivity

Acquired injuries and movement disorders often disrupt the central nervous system across multiple levels, affecting both the brain and spinal cord. While significant progress has been made in functional neuroimaging of the human brain, imaging spinal cord circuits remains a major challenge due to technical limitations. Additionally, the common practice of scanning the brain and spinal cord separately hinders the study of their functional interactions. Recent advancements, including simultaneous spinal cord-brain fMRI technique developed in my lab and others, provide a promising solution to this challenge. In this talk, I will explore the technical hurdles associated with spinal cord neuroimaging and present recent findings that demonstrate the feasibility and value of simultaneous fMRI for investigating brain-spinal cord circuits under task and resting-state conditions. Finally, I will highlight key directions for future research to overcome remaining obstacles and advance the field of spinal cord-brain fMRI.

Konnie Urbaniak, City of Hope

Deciphering HSC Populations in Sickle Cell Disease using Bayesian Network Modeling and Network Theory

  • Authors: Konstancja Urbaniak, Greta Zara, Grigoriy Gogoshin, Andrei Rodin, NadiaCarlesso, Sergio Branciamore

Sickle Cell Disease (SCD) is a hereditary blood disorder characterized by the production of abnormal hemoglobin, leading to deformed erythrocytes and compromised oxygen transport. Despite advancements in therapeutic strategies, SCD remains a global health burden affecting millions worldwide. Hematopoietic stem cell (HSC) transplantation remains the only curative option. Our study aims to investigate the role of HSCs in SCD. Using scRNAseq of bone marrow samples from our SCD patients and healthy donors (HD), we identified distinct HSC populations. While two HSC populations were observed in HD samples, only one population was present in SCD patients. To elucidate the genetic drivers of these differences and explore their relationships, we employed a Bayesian Network using our BNOmics software, an explainable machine learning approach, and network theory. The latter provided insights into the collective behavior of the key drivers in HSC of SCD and their potential implications for disease progression as well as it shed a light on patient Heterogeneity. Our findings demonstrate the power of explainable machine learning techniques to unravel complex cellular relationships at the gene level of HSCs in SCD, paving the way for novel therapeutic insights.
Andrei Rodin, City of Hope

Bowtie architectures: from evolution by gene duplication to artificial neural networks (with application to immune signaling)

Modern deep learning practices and architectures (such as overgrowing and pruning of artificial neural networks, ANNs) have long been designed on foundations of task-specific fit and optimization, as opposed to their original foundations of biomimicry. However, the structure and function of these modern ANNs are often analogous to real-life biological networks. We propose that the ubiquitous information-theoretic and optimization principles underlying the development of ANNs are similar to those guiding the macro-evolution of biological networks and that insights gained from one field can be applied to the other. We look at the evolution and organization of the bowtie network structure of the Janus kinase - signal transducers and activators of transcription (JAK-STAT) pathway, and carry out ANN simulation experiments to demonstrate that an increase in the network's input and output complexity does not necessarily require a more complex intermediate layer. This observation should guide novel biomarker and drug discovery-namely, to prioritize sections of the biological networks in which information is most compressed as opposed to the periphery of the network.