Loading...
Organizing Committee
Case Western Reserve University
Florida State University
Dartmouth College
University of Utah
University of Utah
Abstract
Mathematical models play a vital role in understanding complex biological systems. However, these systems are often characterized by indirectly and partially observed, noisy, and high-dimensional dynamics and data. Current models often fail to adequately account for the various sources of uncertainty within the data, leading to inaccurate or incomplete models with limited explanatory power.
The field of uncertainty quantification (UQ) offers approaches to address these challenges. UQ is the mathematical study of how uncertainty influences and propagates through models, allowing researchers to quantify the impacts of uncertainty on model-based predictions, inference, and design. Despite its success in other scientific domains, the application of UQ techniques in mathematical biology remains underdeveloped.
This workshop capitalizes on the significant potential for mutual advancement in UQ and mathematical biology through cross-disciplinary collaboration. Bringing together UQ practitioners and mathematical biologists will allow UQ methodologies to be applied to explore and improve biological models through computational methods. Conversely, the unique challenges of modeling uncertainty in biological systems can inspire the development of novel UQ techniques. By fostering research collaborations at the intersection of computational and mathematical biology, UQ algorithms, and data-driven learning, this workshop aims to galvanize advancements across these fields.
The field of uncertainty quantification (UQ) offers approaches to address these challenges. UQ is the mathematical study of how uncertainty influences and propagates through models, allowing researchers to quantify the impacts of uncertainty on model-based predictions, inference, and design. Despite its success in other scientific domains, the application of UQ techniques in mathematical biology remains underdeveloped.
This workshop capitalizes on the significant potential for mutual advancement in UQ and mathematical biology through cross-disciplinary collaboration. Bringing together UQ practitioners and mathematical biologists will allow UQ methodologies to be applied to explore and improve biological models through computational methods. Conversely, the unique challenges of modeling uncertainty in biological systems can inspire the development of novel UQ techniques. By fostering research collaborations at the intersection of computational and mathematical biology, UQ algorithms, and data-driven learning, this workshop aims to galvanize advancements across these fields.
Loading...
Loading participant list in background...
Workshop Schedule
Monday, May 05, 2025
8:50 - 9:00 AM EDT
Welcome
11th Floor Lecture Hall
Nathaniel Whitaker, ICERM/ University of Massachusetts Amherst
9:00 - 10:00 AM EDT
Introductions & Overview
Opening Remarks - 11th Floor Lecture Hall
Session Chair
Akil Narayan, University of Utah
Session Chair
Jody Reimer, University of Utah
10:00 - 10:30 AM EDT
Coffee Break
11th Floor Collaborative Space
10:30 - 11:15 AM EDT
Sensitivity Analysis and Uncertainty Quantification for Pharmacokinetic Models
11th Floor Lecture Hall
Speaker
Ralph Smith, North Carolina State University
Session Chair
Linh Huynh, Dartmouth College
Abstract
The quantification of uncertainties inherent to biological and pharmacokinetic models, parameters, and experiments is critical to assess the accuracy of predictions. This presentation will focus on the use of sensitivity analysis and uncertainty quantification techniques to establish the predictive capabilities of selected models. The initial discussion will illustrate how sensitivity analysis and parameter subset selection techniques can be used to isolate model parameters, which are informed by measured data. To demonstrate the role of parameter selection techniques, we consider a minimal physiologically-based pharmacokinetic (mPBPK) model of the brain, which was developed by Bloomingdale, Bakshi, Maass, et al. (2021) to investigate antibody therapeutics. This example will illustrate both the reduction in parameter complexity, which can be achieved, and the dependence of parameter sensitivity on dosage. The presentation will then focus on the use of Bayesian inference techniques to quantify parameter uncertainties in a manner that can be subsequently employed to construct prediction intervals for model responses. This inverse and forward uncertainty analysis will be illustrated in the context of an HIV model. Open questions and future research directions will be noted throughout the presentation.
11:30 AM - 12:15 PM EDT
Quantifying the Uncertainty of Large Language Models Using Spin Glass Theory, with Application to Evolutionary Biology
11th Floor Lecture Hall
Speaker
Linh Huynh, Dartmouth College
Session Chair
Akil Narayan, University of Utah
Abstract
In recent years, Large Language Models (LLMs) have revolutionized Natural Language Processing with their ability to generate human-like texts. However, a fundamental challenge remains in understanding the underlying mechanisms driving their emergent behaviors, particularly the randomness in their outputs. This talk discusses the application of spin glass theory as a mathematical framework to quantify the uncertainty of LLMs. By making connections between LLMs and spin glass models, which are traditionally used in statistical physics and probability to describe disordered networks with random interactions and frustrations (conflicting constraints), we can gain insights into the high-dimensional optimization landscapes of LLMs, the uncertainty in their outputs, and the role of noise in their learning process.
In the first half of the talk, I will discuss my joint work with Matthew Harper on using topology to study the shape of spin glass landscapes, which can help inform the design of optimization algorithms. In the second half, together with my undergraduate students Jackson George, Zach Yusaf, and Stephanie Zoltick, we will showcase our work on LLMs’ temperature-based phase transitions in the context of Flitzing, a unique tradition at Dartmouth College where students send rhymed emails to ask peers out. I will conclude by discussing how this framework can also be applied to studying evolutionary biology.
12:30 - 2:30 PM EDT
Networking & Mentoring Lunch
Working Lunch - 11th Floor Collaborative Space
2:30 - 3:30 PM EDT
Group Work
Assigned Group Work Space
3:30 - 4:00 PM EDT
Coffee Break
11th Floor Collaborative Space
4:00 - 4:45 PM EDT
Bayesian methods for addressing the muscle recruitment problem
11th Floor Lecture Hall
Speaker
Erkki Somersalo, Case Western Reserve University
Session Chair
Nicholas Cogan, Florida State University
Abstract
The musculoskeletal system in the human and mammalian bodies is characterized by a rich redundancy: A given movement such as level walking through recruitment of muscles can be effectuated in infinitely many ways, as the number of muscles exceeds the number of degrees of freedom, in practice, the joint angles, needed to characterize the motion. The uncontrolled manifold hypothesis states that in typical repetitive tasks, the central nervous system controls only few groups of muscles that are organized in a synergetic way, leaving a significant freedom to adjust the movements if needed, e.g., to secure the joints by co-contractions if the external conditions change; Think about walking on a slippery surface. The muscle recruitment problem in biomechanics is to characterize all possible muscle activation configurations corresponding to an observed movement, and to identify the synergies during the task.
Among other things, this helps to get insight into the changed role of muscles in musculo/neurodegenerative states such as Parkinson's disease, and may be useful tool for functional electrical stimulus (FES) applications as well as in treatment of injuries. In this talk, we give an overview of a computational package ""Myobolica"" developed for Bayesian sampling of the muscle activation states, as well as on applications of it.
4:45 - 4:50 PM EDT
Group Photo (Immediately After Talk)
11th Floor Lecture Hall
5:00 - 6:30 PM EDT
Reception
11th Floor Collaborative Space
Tuesday, May 06, 2025
9:00 - 9:45 AM EDT
Efficient Krylov subspace methods for large-scale hierarchical Bayesian inverse problems
11th Floor Lecture Hall
Speaker
Julianne Chung, Emory University
Session Chair
Nicholas Cogan, Florida State University
Abstract
Uncertainty quantification for large-scale inverse problems remains a challenging task. For linear inverse problems with additive Gaussian noise and Gaussian priors, the posterior is Gaussian but sampling can be challenging, especially for problems with a very large number of unknown parameters (e.g., dynamic inverse problems) and for problems where computation of the square root and inverse of the prior covariance matrix are not feasible. Moreover, for hierarchical problems where several hyperparameters that define the prior and the noise model must be estimated from the data, the posterior distribution may no longer be Gaussian, even if the forward operator is linear. Performing large-scale uncertainty quantification for these hierarchical settings requires new computational techniques.
In this work, we consider a hierarchical Bayesian framework and exploit generalized Golub-Kahan based methods to efficiently sample from the posterior distribution. We also estimate the hyperparameters using a maximum a posteriori estimate of the marginalized posterior distribution, where a stochastic average approximation of the objective function and a preconditioned Lanczos method are used to compute efficient approximations of the function and gradient evaluations. We demonstrate the performance of our approach on various static and dynamic tomography problems.
10:00 - 10:30 AM EDT
Coffee Break
11th Floor Collaborative Space
10:30 - 11:15 AM EDT
Quantification of total uncertainty in modeling biological systems
11th Floor Lecture Hall
Speaker
George Karniadakis, Brown University
Session Chair
Nicholas Cogan, Florida State University
Abstract
Quantification of the total uncertainty due to multiple sources includes irreducible aleatoric uncertainty due to noise in the data, epistemic uncertainty induced by insufficient data or inadequate parameterization and model-form uncertainty related to the use of misspecified model equations. We willpresent several examples and focus on a six-compartment stiff ODE system, the CVSim-6 model, developed in the context of human physiology. The system is first analysed by progressively removing data while estimating an increasing number of parameters, and subsequently by investigating total uncertainty under model-form misspecification of nonlinear resistance in the pulmonary compartment.
11:30 AM - 12:30 PM EDT
Panel Discussion
11th Floor Lecture Hall
Session Chair
Nicholas Cogan, Florida State University
Panelist
Daniela Calvetti, Case Western Reserve University
Panelist
Julianne Chung, Emory University
Panelist
Alexander Hoover, Cleveland State University
12:30 - 2:30 PM EDT
Lunch/Free Time
2:30 - 4:30 PM EDT
Mini-Tutorial
Tutorial - 11th Floor Lecture Hall
Session Chair
Daniela Calvetti, Case Western Reserve University
3:30 - 4:00 PM EDT
Coffee Break
11th Floor Collaborative Space
Wednesday, May 07, 2025
9:00 - 9:45 AM EDT
Distributed delay constructively impacts the dynamics of genetic regulatory networks
Virtual
Virtual Speaker
William Ott, University of Houston
Session Chair
Jody Reimer, University of Utah
Abstract
Delay is an inherent feature of genetic regulatory networks. It represents the time required for the assembly of functional regulator proteins. The protein production process is complex, as it includes transcription, translocation, translation, folding, and oligomerization. Because these steps are noisy, the resulting delay associated with protein production is distributed (random). My collaborators and I have shown that distributed delay can significantly impact the dynamics of genetic regulatory networks in constructive ways. It can accelerate signaling in feed-forward architectures, denoise biochemical oscillators, and stabilize metastable states in bistable systems. I will discuss these results and then examine efforts to infer delay distributions from fluorescence microscopy data. I will conclude by posing research problems that may be amenable to techniques from uncertainty quantification.
10:00 - 10:30 AM EDT
Coffee Break
11th Floor Collaborative Space
10:30 - 11:15 AM EDT
Sequential Bayesian Inference for Uncertainty Quantification of Biological Systems
11th Floor Lecture Hall
Speaker
Andrea Arnold, Worcester Polytechnic Institute
Session Chair
Jody Reimer, University of Utah
Abstract
Sequential Bayesian inference (SBI) offers an attractive data assimilation framework for uncertainty quantification of biological systems with temporal dynamics. In this talk, I will give an overview of SBI methods and their use in solving nonstationary inverse problems characterized by dynamical systems, emphasizing novel applications in biology and biomedical engineering, including recent work identifying tissue physical properties for laser surgery.
11:30 AM - 12:30 PM EDT
Group Work
Assigned Group Work Space
12:30 - 2:30 PM EDT
Lunch/Free Time
2:30 - 4:00 PM EDT
4:00 - 5:00 PM EDT
Group Work
Assigned Group Work Space
Thursday, May 08, 2025
9:00 - 9:45 AM EDT
Data-Driven Methods for Two Inference Problems in Mathematical Biology via Optimal Transport
11th Floor Lecture Hall
Speaker
Wenjun Zhao, UBC
Session Chair
Akil Narayan, University of Utah
Abstract
In this talk, I will present two inference problems arising in dynamical systems in mathematical biology and discuss data-driven numerical methods based on optimal transport. First, we revisit the problem of inferring regulatory networks from time-stamped single-cell transcriptomics data. Optimal transport has been widely used to infer ancestor-descendant relationships across time points; we leverage this structure to infer gene regulatory interactions using Granger causality. Second, time permitting, I will briefly discuss bifurcation tracing in models of pattern formation, including reaction-diffusion systems and agent-based models.
10:00 - 10:30 AM EDT
Coffee Break
11th Floor Collaborative Space
10:30 - 11:15 AM EDT
Computationally efficient uncertainty quantification for sparse Bayesian learning
11th Floor Lecture Hall
Speaker
Jan Glaubitz, Linköping University
Session Chair
Akil Narayan, University of Utah
Abstract
Sparse Bayesian learning (SBL) is an advanced statistical modeling technique for inverse problems that builds upon traditional Bayesian methods by integrating hierarchical structures within prior distributions. This approach allows for extracting intricate relationships between parameters at various levels, fostering information sharing throughout the model. It is particularly effective when dealing with limited, noisy, or indirect data, yielding more accurate and robust inferences. Consequently, SBL has proven successful in diverse fields, such as machine learning, signal processing, remote sensing, and mathematical biology.
In this talk, we introduce a novel approach for efficient Markov chain Monte Carlo (MCMC) sampling from the complex SBL posterior. The core innovation involves using prior-normalizing transport maps, which are deterministic couplings that transform the hierarchical sparsity-promoting SBL prior into a standard normal distribution. These maps transform the complex target posterior into a simpler reference distribution equipped with a standard normal prior that can be sampled more efficiently. Numerical experiments will demonstrate order-of-magnitude speedups for standard MCMC techniques.
This talk is based on joint work with Youssef Marzouk (MIT).
11:30 AM - 12:15 PM EDT
Uncertainty Quantification in Quantitative Systems Pharmacology (QSP): An illustrative example applied to microbiome therapies
11th Floor Lecture Hall
Speaker
Marissa Renardy, GSK
Session Chair
Akil Narayan, University of Utah
Abstract
Quantitative Systems Pharmacology (QSP) integrates biological knowledge of disease pathology and drug mechanism of action into a mathematical or computational framework to simulate patient responses to drug treatment. QSP is used to support decision making across all stages of drug development, and has been increasingly used in regulatory submissions since 2013. Due to the heterogeneous nature of disease and treatment response, and because of unknown aspects of the relevant biology, uncertainty quantification is critical to meaningfully interpret QSP results and predictions. In this talk, I will give a brief introduction of the field of QSP, applications of uncertainty quantification in QSP such as sensitivity analysis and virtual populations, and will show an illustrative example applied to a live biotherapeutic product (LBP). In this example, we developed a novel cellular kinetic-pharmacodynamic quantitative systems pharmacology model of an LBP targeting the gut microbiome. The model is calibrated and validated to published data from healthy volunteers, then used to simulate the impact of treatment dose, frequency, and duration as well as antibiotic pretreatment on a key therapeutic metabolite.
12:30 - 2:30 PM EDT
Lunch/Free Time
2:30 - 3:30 PM EDT
Group Work
Assigned Group Work Space
3:30 - 4:00 PM EDT
Coffee Break
11th Floor Collaborative Space
4:00 - 5:00 PM EDT
Group Work
Assigned Group Work Space
Friday, May 09, 2025
9:00 - 9:45 AM EDT
Uncertainty quantification in cardiovascular modeling
11th Floor Lecture Hall
Speaker
Mette Olufsen, North Carolina State University
Session Chair
Daniela Calvetti, Case Western Reserve University
Abstract
Cardiovascular system dynamics have long been studied using mathematical models ranging from simple algebraic models, over compartment models to 3D fluiddynamics and mechanics models. With the uptake in availability of data, the need for predictive models and the development of digital twins is rising. This talk will provide an overview of some of the many sources of uncertainty in these models. We compare use results from sensitivity analysis for asymptotic prediction of confidence and prediction intervals with intervals determined using Bayesian sampling. We also address how to incorporate model mismatch using Gaussian Processes and demonstrate the physics based neural network model. These topics will be demonstrated in a closed loop model used to understand the importance of ventricular-ventricular interaction in patients with pulmonary hypertension and in a 1D fluid dynamics network model developed to study what vessels to open in patients with chronic thromboembolic pulmonary hypertension.
10:00 - 10:30 AM EDT
Coffee Break
11th Floor Collaborative Space
10:30 - 11:15 AM EDT
Bayesian information-theoretic approach to determine effective scanning protocols of cancer patients
11th Floor Lecture Hall
Speaker
Heyrim Cho, Arizona State University
Session Chair
Daniela Calvetti, Case Western Reserve University
Abstract
The aspect of limited temporal data is one of the many challenges when dealing with clinical data. The amount of data that can be practically collected in everyday patients during the therapy is very limited due to the financial cost and the patient’s burden. This motivates us to transfer the mathematical and computational models to meet the challenges in clinical data, before we use them to guide patient therapy via prediction. In this talk, I will discuss modeling approaches to tackle this problem. For instance, I will discuss a Bayesian information-theoretic approach to determine effective scanning protocols of cancer patients. We propose a modified mutual information function with a temporal penalty term to account for the loss of temporal data. The effectiveness of our framework is demonstrated in determining scanning scheduling for radiotherapy and androgen suppression therapy patients.
11:30 AM - 12:15 PM EDT
Closing Remarks
11th Floor Lecture Hall
Session Chair
Daniela Calvetti, Case Western Reserve University
Request Reimbursement
This section is for general purposes only and does not indicate that all attendees receive funding. Please refer to your personalized invitation to review your offer.
- ORCID iD
- As this program is funded by the National Science Foundation (NSF), ICERM is required to collect your ORCID iD if you are receiving funding to attend this program. Be sure to add your ORCID iD to your Cube profile as soon as possible to avoid delaying your reimbursement.
- Acceptable Costs
- 1 roundtrip between your home institute and ICERM
- Flights on U.S. or E.U. airlines – economy class to either Providence airport (PVD) or Boston airport (BOS)
- Ground Transportation to and from airports and ICERM.
- Unacceptable Costs
- Flights on non-U.S. or non-E.U. airlines
- Flights on U.K. airlines
- Seats in economy plus, business class, or first class
- Change ticket fees of any kind
- Multi-use bus passes
- Meals or incidentals
- Advance Approval Required
- Personal car travel to ICERM from outside New England
- Multiple-destination plane ticket; does not include layovers to reach ICERM
- Arriving or departing from ICERM more than a day before or day after the program
- Multiple trips to ICERM
- Rental car to/from ICERM
- Flights on a Swiss, Japanese, or Australian airline
- Arriving or departing from airport other than PVD/BOS or home institution’s local airport
- 2 one-way plane tickets to create a roundtrip (often purchased from Expedia, Orbitz, etc.)
- Travel Maximum Contributions
- New England: $350
- Other contiguous US: $850
- Asia & Oceania: $2,000
- All other locations: $1,500
- Note these rates were updated in Spring 2023 and supersede any prior invitation rates.
- Reimbursement Requests
Request Reimbursement with Cube
Refer to the back of your ID badge for more information. Checklists are available at the front desk and in the Reimbursement section of Cube.
- Reimbursement Tips
- Scanned original receipts are required for all expenses
- Airfare receipt must show full itinerary and payment
- ICERM does not offer per diem or meal reimbursement
- Allowable mileage is reimbursed at prevailing IRS Business Rate and trip documented via Google Maps PDF
- Keep all documentation until you receive your reimbursement!
- Reimbursement Timing
6–8 weeks after all documentation is sent to ICERM. All reimbursements are reviewed by multiple Brown University offices.
- Reimbursement Deadline
Submissions must be received within 30 days of ICERM departure to avoid applicable taxes. No submissions are accepted after six months.
Projects
Data-informed model reduction for inference and prediction from non-identifiable models
Matthew Simpson (Queensland University of Technology)
Many mathematical models in the field of theoretical biology involve challenges relating to parameter identifiability. Non-identifiability implies that different combinations of parameter values lead to indistinguishable solutions of the mathematical model. This means that it is difficult, and sometimes impossible, to explain the mechanistic origin of observations using a non-identifiable mathematical model. A standard approach to deal with structurally non-identifiable models is to use reparameterisation, which typically focuses on the structure of the mathematical model without accounting for the impact of noisy, finite data.
In this workshop we will explore a simple computational approach for model reduction, via likelihood reparameterisation, that can be applied to both structurally non-identifiable and practically non-identifiable problems. We will construct simplified, identifiable mathematical models that enable model-based predictions for a range of continuum models based on different classes of commonly-used differential equations. Through a series of computational experiments, we will illustrate how to deal with a range of noise models that relate the solution of the mathematical model with noisy observations. A key focus is to illustrate how computationally efficient model-based predictions can be made from reduced models.
In this workshop we will explore a simple computational approach for model reduction, via likelihood reparameterisation, that can be applied to both structurally non-identifiable and practically non-identifiable problems. We will construct simplified, identifiable mathematical models that enable model-based predictions for a range of continuum models based on different classes of commonly-used differential equations. Through a series of computational experiments, we will illustrate how to deal with a range of noise models that relate the solution of the mathematical model with noisy observations. A key focus is to illustrate how computationally efficient model-based predictions can be made from reduced models.
Surrogate Modeling for Scalable Uncertainty Quantification in ABMs of Cancer Growth
Daniel Bergman (University of Maryland), Harsh Jain (University of Minnesota Duluth)
Agent-based models (ABMs) have become an indispensable tool for studying cancer growth and treatment response, as they naturally capture the complexity, heterogeneity, and spatial interactions characteristic of tumors. However, the computational costs associated with realistic ABM simulations often make uncertainty quantification (UQ)—including global sensitivity analysis (GSA) and inverse UQ—prohibitively expensive.
To overcome these computational challenges, our working group will explore the use of surrogate modeling techniques. Surrogate models serve as computationally efficient proxies for the original ABM, enabling rapid evaluation and thus facilitating extensive uncertainty analyses. We propose specifically to evaluate and compare direct UQ approaches, machine learning-based surrogate methods, and our recently developed method, Surrogate Modeling for Recapitulating Global Sensitivity (SMoRe GloS). SMoRe GloS explicitly preserves biological interpretability while substantially reducing computational burden.
Our initial focus will be on global sensitivity analysis (GSA). However, methods developed through this working group will be broadly applicable to other areas of UQ, such as parameter estimation and inverse modeling. We will use the open-source PhysiCell platform—a flexible, C++-based modeling framework widely adopted in cancer modeling—as our primary testbed. Developed with outreach and education as core missions alongside power, extensibility, and cross-platform compatibility, PhysiCell is readily accessible even to participants new to ABMs. Furthermore, a GUI interface enables complex modeling without C++ coding. A PhysiCell core developer will also be on the team to help members get set up and answer questions. Additionally, we encourage participants to bring their own ABMs or biological applications to the group, enriching the discussion and providing diverse case studies for method evaluation.
To overcome these computational challenges, our working group will explore the use of surrogate modeling techniques. Surrogate models serve as computationally efficient proxies for the original ABM, enabling rapid evaluation and thus facilitating extensive uncertainty analyses. We propose specifically to evaluate and compare direct UQ approaches, machine learning-based surrogate methods, and our recently developed method, Surrogate Modeling for Recapitulating Global Sensitivity (SMoRe GloS). SMoRe GloS explicitly preserves biological interpretability while substantially reducing computational burden.
Our initial focus will be on global sensitivity analysis (GSA). However, methods developed through this working group will be broadly applicable to other areas of UQ, such as parameter estimation and inverse modeling. We will use the open-source PhysiCell platform—a flexible, C++-based modeling framework widely adopted in cancer modeling—as our primary testbed. Developed with outreach and education as core missions alongside power, extensibility, and cross-platform compatibility, PhysiCell is readily accessible even to participants new to ABMs. Furthermore, a GUI interface enables complex modeling without C++ coding. A PhysiCell core developer will also be on the team to help members get set up and answer questions. Additionally, we encourage participants to bring their own ABMs or biological applications to the group, enriching the discussion and providing diverse case studies for method evaluation.
Connections between identifiability and uncertainty quantification (in dynamic systems biology modeling)
Alejandro Villaverde (Universidade de Vigo)
The proposed research topic concerns the relationship between the analysis of (structural) identifiability and observability (SIO), on the one hand, and uncertainty quantification (UQ), on the other. Both concepts are related to uncertainty, but in different ways. The SIO properties are fully determined by the model equations, they do not depend on the data [Wieland et al., 2021; Villaverde, 2019]. UQ, on the other hand, is strongly dependent on the available data. Some authors (e.g., [Norden et al.]) have argued that SIO is related to epistemic uncertainty and UQ to aleatoric uncertainty. SIO may be analyzed with symbolic approaches such as differential algebra [Dong et al., 2023] and differential geometry [Díaz-Seoane et al., 2023]. Another possibility is to use simulation-based numerical approaches via sensitivities [van Willigenburg et al. 2022, Joubert et al. 2020] or optimization, using the profile likelihood approach (PL) [Kreutz et al., 2013].
In this project we propose to explore the links between SIO and UQ in several directions.
As modeling framework we will consider dynamic models of deterministic systems in continuous time (and their corresponding hybrid extensions, in the case of item III). Most work in this area has been done in ordinary differential equations (ODEs), although many concepts and ideas can also be applied to partial differential equation systems (PDEs). The range of possible biological applications is very wide, including essentially every biosystem or bioprocess that can be modelled in this way (see, e.g., the recent/current special issue of the Philos. Trans. R. Soc. A).
In this project we propose to explore the links between SIO and UQ in several directions.
As modeling framework we will consider dynamic models of deterministic systems in continuous time (and their corresponding hybrid extensions, in the case of item III). Most work in this area has been done in ordinary differential equations (ODEs), although many concepts and ideas can also be applied to partial differential equation systems (PDEs). The range of possible biological applications is very wide, including essentially every biosystem or bioprocess that can be modelled in this way (see, e.g., the recent/current special issue of the Philos. Trans. R. Soc. A).
Quantitative Systems Pharmacology virtual populations simulations for oncology drug development: evaluating the problem space through the lens of uncertainty quantification
Blerta Shtylla (Pfizer)
Simulation and analysis of virtual populations are increasingly being used to guide decisions in drug development within the context of model informed drug development (MIDD). While there are a multitude of algorithmic approaches for selecting virtual populations, most rely on smart sampling of model parameter values to allow a mechanistic quantitative systems pharmacology (QSP) model to match the diverse responses observed in real clinical populations [1,2]. We will review key existing methodologies and computational approaches to address specific challenges in solid-tumor oncology for matching tumor size and progression free survival data. We survey in more detail a recently published case study from our group focusing on anaplastic lymphoma kinase inhibitors (ALKi), which have shown great promise in circumventing on-target resistance mutations in ALK+ non-small cell lung cancer (NSCLC) [3]. A virtual population strategy was outlined that focused on capturing the range of resistance to ALKi therapies that emerge. Comprehensive sampling of potential resistance mechanisms was essential for understanding clinical response to current therapies.
Using the case study recently published in [3], I will outline some open challenges for improving the efficiency and quality of virtual populations. For efficiency, we will be interested in exploring methods that extend approaches outlined in [2] with an eye on whether computational efficiency can be increased through integrated use of parameter identifiability sampling algorithms. The goal is to improve how large dimensional parameter spaces can be sampled both in the context of computational efficiency as well as virtual patient heterogeneity. Questions regarding quality are related to, for example, how many virtual patients are needed to fully capture the potential mechanistic variability implied by the model and the data at hand. Answering these questions requires metrics that account for the interaction between prior parameter bounds, non-identifiability, and population variability within the sampling procedure.
Using the case study recently published in [3], I will outline some open challenges for improving the efficiency and quality of virtual populations. For efficiency, we will be interested in exploring methods that extend approaches outlined in [2] with an eye on whether computational efficiency can be increased through integrated use of parameter identifiability sampling algorithms. The goal is to improve how large dimensional parameter spaces can be sampled both in the context of computational efficiency as well as virtual patient heterogeneity. Questions regarding quality are related to, for example, how many virtual patients are needed to fully capture the potential mechanistic variability implied by the model and the data at hand. Answering these questions requires metrics that account for the interaction between prior parameter bounds, non-identifiability, and population variability within the sampling procedure.
Modeling immune mechanisms of inflammatory exacerbations
Marissa Renardy (GSK)
Chronic obstructive pulmonary disease (COPD) is a chronic condition caused by damage to the lungs, which leads to inflammation and other problems that obstruct airflow. COPD affects more than 14 million adults in the US. COPD includes both emphysema and chronic bronchitis, and people with COPD often have a mixture of both. The repeated and progressive activation of immune cells plays a pivotal role in the pathogenesis of COPD [1]. A COPD exacerbation is an acute worsening of respiratory symptoms that may last for days or weeks, often triggered by a viral or bacterial infection or exposure to pollutants, which can leave behind permanent lung damage and accelerate disease progression [2]. Thus, decreasing exacerbations is a key goal of therapeutics. Due to the heterogeneous and random nature of exacerbations, predicting exacerbation frequency and severity is difficult. Quantitative approaches include machine learning [3], regression models [4], biomechanical models [5,6], and mechanistic immunological models [7].
The goal of this project is to develop a methodology for simulating inflammatory exacerbations in COPD, to be applied to a mechanistic model of immune interactions in order to predict or understand (1) what features are predictive of exacerbation frequency/severity, and (2) what immunological mechanisms are most effective to target (and under what conditions) to reduce exacerbation frequency/severity. Uncertainty quantification approaches will be critical in capturing the impact of patient heterogeneity and randomness of exacerbations.
The goal of this project is to develop a methodology for simulating inflammatory exacerbations in COPD, to be applied to a mechanistic model of immune interactions in order to predict or understand (1) what features are predictive of exacerbation frequency/severity, and (2) what immunological mechanisms are most effective to target (and under what conditions) to reduce exacerbation frequency/severity. Uncertainty quantification approaches will be critical in capturing the impact of patient heterogeneity and randomness of exacerbations.
Uncertainty Quantification in Physiological Models
Mette Olufsen (North Carolina State University), Mitchel Colebank (University of South Carolina)
Uncertainty quantification (UQ) is an essential analysis tool for employing and calibrating mathematical models to clinical data, yet many modeling studies do not address uncertainty in measurements, model structure, or parameter estimates. Often, data measured include a combination of static, spatial, and timeseries data, some of which are often incomplete. Gaining more insight into how to assess uncertainty is essential for the generation of cardiovascular digital twins and using models to extract biomarkers that can be essential for diagnosis and treatment planning. We will discuss three modeling applications: (1) closed-loop cardiovascular compartment (ODE) models informed by different data modalities, (2) distributed one-dimensional (1D) fluid dynamic network models informed in part by medical imaging geometric data, and (3) multiscale/multisystem models which combines multiple spatial or temporal scales (e.g., growth and remodeling over months) or multiple physiologic systems (e.g., autonomic control or inflammatory signaling)