Special Session 106: Data-Driven Multiscale Modeling and Model Reduction Techniques in Biomedicine: Bridging Scales and Complexity

Data-Driven Models for Extended Reality Solutions in General Anesthesia Management.

Ghada Ben Othman Ghent University Belgium Co-Author(s):    Ghada BEN OTHMAN, Clara Mihaela IONESCU

Abstract: This presentation introduces an extension to system theory as a novel framework for modeling clinical data under significant uncertainty and poor identifiability conditions, common challenges in medical systems. These challenges arise from ethical, safety, and regulatory constraints, limiting the persistent drug-related excitation of the human body. Moreover, the drug-dose effect relationship is complicated by substantial inter- and intra-patient variability. The absence of suitable instrumentation for direct measurement, relying instead on inferences and surrogate metrics, adds further complexity. The efficacy of our approach was examined in clinical data from patients monitored during the induction phase of target-controlled intravenous anesthesia. The proposed method delivered models with physiological explainable parameters and suitable for closed-loop control of anesthesia. A notable advantage of this approach is its robustness in the face of uncertainty. The work is the first piece of the puzzle towards Extended Reality solutions encompassing virtual, augmented, and mixed reality in general anesthesia management.

Mechanisms of cancer invasion and progression: insights from agent-based models

Andreas Deutsch Centre for Interdisciplinary Digital Sciences, Dresden University of Technology Germany Co-Author(s):

Abstract: Cancer invasion may be viewed as collective phenomenon emerging in populations of normal and malignant cells. As such it can be studied with agent-based models, e.g. cellular automata. I will provide examples of such models to analyze breast and glioma invasion as well as the emergence of phenotypic heterogeneity due to cellular interactions in growing tumors. Furthermore, I will present models which shed light on cancer progression.

COVID-19 in Greece: the dynamics of the 4th, 5th and 6th waves

Dimitris A Goussis Khalifa University United Arab Emirates Co-Author(s):

Abstract: The results of an alternative methodology for making predictions about the COVID-19 pandemic are presented. The methodology is based on the Computational Singular Perturbation method, which delivers algorithmically singular perturbation analysis results for complex multiscale systems. Instead of focusing on the various population profiles (subjected to instabilities introduced by the fitting process), this methodology focuses on the time scale that characterizes the intensity and duration of the outbreak phase. The analysis leads to the conclusion that the prediction of the inflection point in the profile of the active cases is much more robust and accurate than the prediction of the. Since the inflection point precedes the peak, this methodology can serve as an early warning of the peak. In addition, CSP diagnostics are shown to provide significant information regarding the paths between the various populations (healthy, exposed, infected, etc) that contribute the most to the outbreak phase. The accuracy of the new methodology in analyzing pandemic outbreaks is demonstrated for the 4th, 5th and 6th outbreak waves of COVID-19 in Greece.

How to make clinical predictions when we do not know everything? Synergies between dynamic modelling and AI

Haralampos Hatzikirou Khalifa University United Arab Emirates Co-Author(s):

Abstract: In clinical practice, a plethora of examinations is conducted to assess the state of a certain pathology. These span from blood sample analysis, clinical imaging (e.g. CT, MRI) and biopsy sampling are among the most important diagnostic and prognostic tools. Such medical data correspond to snapshots in time of the patient`s state, since current standard of care (SoC) is not based on emergent technologies of real-time measurements, such as liquid biopsies or biosensors. Moreover, clinical data refer to different biological scales since imaging, such as MRI, typically provides an organ level picture of a disease (macroscopic), biopsies represent cellular patterns at a tissue (mesoscopic) level and -omics, FACS or molecular markers allow for sub-cellular insights. Finally, the biophysical mechanisms that regulate phenomena in all these scales are not completely known. Therefore, current clinical care faces the following challenges: (C1) data collection is sparse in time since it relies on patient`s clinical presentation, (C2) we lack the knowledge/uncertainty of the mechanisms involved in regulating these data variables across different scales (structural uncertainty), and (C3) medical data are multiscale. Therefore, integrating these data to predict the future of a disease and propose an appropriate treatment is a formidable task. I propose to harness the ability of mechanistic models to integrating the existing biological knowledge and deal with the emerging dynamics. At the same time complete the missing knowledge by using data intensive techniques. Here I will present (i) a Bayesian regression framework of combining models and machine learning to predict tumor growth and (ii) model-driven classification method to assess the graft loss risk in kidney transplantation patients.

Integration of topological data analysis and equation free methods on describing dynamics on random networks

Nikos Kavallaris Karlstad University Sweden Co-Author(s):

Abstract: This paper integrates topological data analysis (TDA) with the equation-free method (EFM) to analyze complex neuronal network dynamics. We map network activity onto a circle and use TDA to identify the minimum radius r for Betti 1 emergence. By employing simulated annealing and a lifting method, we recover network dynamics with fewer nodes, focusing on minimum radious to describe macroscopic behavior. Finally, we perform numerical bifurcation and stability analyses on these dynamics.

Spatio-temporal Dynamics of MMK4 Function for JNK Pathway from Analog to Digital Converter in Response to Stress Intensities

Nuha Loling Othman Osaka University Japan Co-Author(s):    Hisashi Moriizumi, Mutsuhiro Takekawa, Takashi Suzuki

Abstract: Spatio-temporal regulation of mitogen-activated protein kinase (MAPK) signaling is essential for mammalian cells` growth, differentiation, and survival. There are three mitogens involved in response to proliferation, apoptosis, and cytokine production, which are called ERK, p38, and c-JUN N-terminal kinases (JNK). MAPK pathway cascade is a central signal transduction pathway activated by growth factors in response to stimuli by phosphorylating and activating mitogen-activated protein kinase kinase kinase (MAPKKK) - mitogen-activated protein kinase kinase (MAPKK) - MAPK in a stepwise and continuous manner, transmitting signals from upstream to downstream. The activation of MAPK which activates several nuclear targets such as mitogen-activated protein kinase kinase 4 (MKK4) which binds to JNK leads to their translocation from cytoplasm to nucleus. MKK4 is a member of MAPKK family, phosphorylates, and activates JNK in response to cellular stresses and proinflammatory cytokines. JNK which also belongs to the family of mitogen-activated protein kinase (MAPK) has been implicated in the apoptotic response of cancer cells exposed to stress while JNK is required to induce apoptosis. Here, we model the relation of MAPK pathway response to stress intensities for apoptosis in programmed cell death in cancer.

Modelling Post-Operative Glioblastoma Relapse

Andrei Macarie University of Dundee Scotland Co-Author(s):

Abstract: In this work we aim to explore oedema infiltration and predict relapse patterns of GBM. To address this, we propose a novel multiscale mathematical modelling framework to simulate and predict tumour growth, oedema infiltration, and treatment response under various conditions. Simulation results obtained by exploring a large space of post-operatory residual oedema cell distributions led us to formulate the hypothesis that a higher concentration of tumour cells remaining near the surgical cavity edge led to slower and more localized tumour growth. Based on this simulations-inspired hypothesis we explore the ways of reconstructing the personalised initial tumour distribution within the oedema from existing MRI patient data in an inverse problem approach, with the ultimate goal of achieving prediction abilities for our modelling framework. The prediction abilities acquired by our framework through this inverse problem approach are promising, which for instance enabled us to achieve realistic prediction (i.e., match MRI data) of 881 days post-treatment GBM relapse evolution [4]. While further analytical investigations are ongoing, this innovative approach holds promise for reconstructing tumor relapses from readily available clinical data, offering new insights into GBM progression and treatment response.

Data-Driven Identification of Regions for Model Reduction in Multiscale Biomedical Data

Dimitris M Manias Khalifa University United Arab Emirates Co-Author(s):

Abstract: The rapidly increasing volume and complexity of biomedical data, characterized by its multiscale and multidimensional nature, present significant opportunities and challenges for effective model development. Multiscale systems in biomedicine often exhibit complex dynamics that cannot be adequately captured by traditional modeling approaches, thereby necessitating novel data-driven model reduction techniques. Furthermore, the sparse, heterogeneous, and often incomplete nature of biomedical datasets adds layers of complexity, underscoring the need for advanced methodologies capable of extracting meaningful patterns and representing system dynamics across multiple scales. In this work, we present a systematic methodology for identifying regions within the data where reduced models can be effectively constructed, leveraging tools from the Computational Singular Perturbation (CSP) method. By segmenting the data and pinpointing distinct regions of dynamic behavior, our approach facilitates the construction of region-specific reduced models that accurately represent the local properties of the system. This strategy is particularly effective in addressing the inherent multiscale complexity of biological systems, allowing for the development of reduced models that are computationally efficient while retaining the critical features of the underlying dynamics. The proposed methodology employs data-driven techniques to approximate key components, such as the Jacobian matrix, which plays a vital role in identifying timescale separations and dominant subprocesses in the data. By integrating CSP tools, we ensure a systematic partitioning of the data into regions where different reduced models can be reliably constructed, thereby enhancing both the validity and applicability of these models. After identifying the correct data partitioning, we can apply existing numerical methods to reconstruct the corresponding governing equations of the reduced models. This structured approach bridges the gap between data sparsity, multiscale interactions, and model complexity, ultimately contributing to the development of accurate, patient-specific models in the biomedical domain.

Reinforcement learning informs optimal treatment strategies to limit antibiotic resistance

Jacob G Scott Cleveland clinic USA Co-Author(s):

Abstract: Antimicrobial resistance was estimated to be associated with 4.95 million deaths worldwide in 2019. It is possible to frame the antimicrobial resistance problem as a feedback-control problem. If we could optimize this feedback-control problem and translate our findings to the clinic, we could slow, prevent, or reverse the development of high-level drug resistance. Prior work on this topic has relied on systems where the exact dynamics and parameters were known a priori. In this study, we extend this work using a reinforcement learning (RL) approach capable of learning effective drug cycling policies in a system defined by empirically measured fitness landscapes. Crucially, we show that it is possible to learn effective drug cycling policies despite the problems of noisy, limited, or delayed measurement. Given access to a panel of 15 -lactam antibiotics with which to treat the simulated Escherichia coli population, we demonstrate that RL agents outperform two naive treatment paradigms at minimizing the population fitness over time. We also show that RL agents approach the performance of the optimal drug cycling policy. Even when stochastic noise is introduced to the measurements of population fitness, we show that RL agents are capable of maintaining evolving populations at lower growth rates compared to controls. We further tested our approach in arbitrary fitness landscapes of up to 1,024 genotypes. We show that minimization of population fitness using drug cycles is not limited by increasing genome size. Our work represents a proof-of-concept for using AI to control complex evolutionary processes.