Creation of predictive cardiac excitation models at the tissue scale with machine learning in augmented state space

Kabus, D.; de Vries, A. A. F.; Pijnappels, D. A.; Dierckx, H.

doi:10.1101/2023.05.17.540314

Home/Archive/bioRxiv

bioRxiv · 10.1101/2023.05.17.540314

Creation of predictive cardiac excitation models at the tissue scale with machine learning in augmented state space

Kabus, D.·de Vries, A. A. F.·Pijnappels, D. A.·Dierckx, H.

View original source ↗Export metadata ↓

Abstract

The quick and easy creation of fast in-silico models of excitable media is, for instance, needed for patient-specific predictions in diagnostics and decision making in cardiac electrophysiology. We here present a model creation pipeline that not only generates new models quickly, but also only requires data from one easily measurable spatio-temporal variable. These data may, for instance, be an optical voltage mapping recording of the electrical waves in cardiac muscle tissue controlling the heart beat. We use exponential moving averages and compute standard deviations to extract additional states from this one variable to span a sparse discretised state space. The standard deviation in a neighbourhood can be used as a proxy for the gradient. To this augmented state space, we fit a simple polynomial model to predict the evolution of this one state variable. For optical voltage mapping data of human atrial myocyte monolayers electrically stimulated by stochastic burst pacing, the data-driven model is able to describe the excitation and recovery of the system, as well as wave propagation. The data-driven model is also able to predict spiral waves only based on data from focal waves. In contrast to conventional models, with our model creation pipeline new models can be generated in a matter of hours from experiment to fitting, rather than months or years. HighlightsO_LIModels of excitation waves can be created in minutes from experiment to fitting. C_LIO_LIOne variable in space and time is sufficient to create a working excitation model. C_LIO_LIA polynomial can predict excitation waves based on useful extracted features. C_LIO_LISpiral waves in heart muscle tissue can be predicted from focal wave data. C_LI Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=87 SRC="FIGDIR/small/540314v2_ufig1.gif" ALT="Figure 1"> View larger version (22K): org.highwire.dtl.DTLVardef@2165acorg.highwire.dtl.DTLVardef@9a1bfeorg.highwire.dtl.DTLVardef@1a3cd50org.highwire.dtl.DTLVardef@fc5698_HPS_FORMAT_FIGEXP M_FIG C_FIG

Explore related subjects

systems biology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Kabus, D., de Vries, A. A. F., Pijnappels, D. A., Dierckx, H.. 2023-05-19. Creation of predictive cardiac excitation models at the tissue scale with machine learning in augmented state space. https://doi.org/10.1101/2023.05.17.540314

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Large language models complement established workflows for mechanistic model construction and refinement in systems biology

Mechanistic models provide a framework for describing, understanding, and predicting biological processes, but their development requires an interdisciplinary team with expertise in biology, mathematics, and computation. Here we present Octomod, a multi-agent workflow based on large language models (LLMs) that integrates biological knowledge retrieval, model construction, parameter estimation, and hypothesis testing. Built on established software tools, including AMICI and pyPESTO, and community standards, including SBML and PEtab, Octomod supports model development and refinement through natural-language interaction with a user. The agents' proposals must pass defined checks and the user's decision before they are used. Every input, model, check, and decision is recorded, so that model building remains transparent and reproducible. We applied Octomod to four problems. Three used published data and differ in model size and in how much the user guided the run: Octomod constructed a model that matched a published model without user guidance, supported the interactive refinement of a gene-regulatory model, and built models of cell-cell communication that explained the responses of five immune cell types to different stimuli. In the fourth application, on newly generated and unpublished lipidomics data, Octomod helped to determine which processes the measurements could inform, identify unresolved label fluxes, and reduce a supplied model to those processes. Together, these results show how LLMs can strengthen and complement established workflows for constructing and refining biologically informed models.

systems biology↗

bioRxiv2026

Dynamic rewiring of super-enhancers shapes condition-specific genome regulation

Enhancers and promoters communicate through three-dimensional chromatin contacts to control cell13 type-specific gene expression. While chromosome conformation capture has enabled genome-wide mapping of these contacts, and graph-based analyses of the resulting interaction networks have shown that super enhancers occupy highly connected positions. However, how the network positions of super enhancers and typical enhancers changes as cells transition between states remains unclear. In this study, enhancer-promoter networks from high-resolution Hi-C data were constructed, and network18 based approaches were applied to characterize the regulatory architecture and its remodeling during differentiation, using Hi-C and HiChIP datasets from three differentiation systems: PMA-induced K562 differentiation, cardiomyocyte differentiation of h9 cells, and differentiation of primary foreskin keratinocytes. The networks are modular, with the largest clusters consisting of promoters and both enhancer types. Super-enhancer constituents and typical enhancers occupy distinct topological positions. During differentiation, network structure is mainly reshaped by enhancer turnover rather than by enhancer reorganization. Among the persistent enhancers, super-enhancer constituents rewire more extensively than typical enhancers. Functionally, remodeled super-enhancers connect to promoters controlling cell type's identity-defining programs, whereas typical enhancers connect to promoters involved in general metabolic and stress-response loci. These results provide evidence that super-enhancers act as a sensitive, identity-focused layer of chromatin communication. The present study also providesquantitative, network-level evidence for how super-enhancers shape the architecture of chromatin communication.

systems biology↗

bioRxiv2026

A Mathematical Model Connecting Appetite-Driven Food Intake to Metabolic Changes across Diet and GLP-1RA Treatments in Mice

Mouse models are widely used to study obesity and the effects of glucagon-like peptide-1 receptor agonists (GLP-1RAs) on appetite-driven meal responses and eating behaviours. However, connecting these behavioural changes to disease prevention remains limited. One reason is that eating behaviour is often summarised as daily or weekly energy intake, obscuring meal-level dynamics. Here we present a mathematical model of mouse appetite-driven food intake across diets and treatment with the GLP-1RAs semaglutide and exenatide. The model jointly described three studies and passed both qualitative assessments and a statistical {chi}2 test (disagreement<threshold, 348<1111). Predictive performance was subsequently validated using four pre-existing studies and one new study. Finally, the model was used to simulate mouse populations under dietary and GLP-1RA interventions, with predictions showing qualitative agreement with the literature. This work aids preclinical drug discovery, supports mouse study design, and improves understanding of how individual meals accumulate into long-term physiological changes.

systems biology↗