Data-driven discovery of dynamical models in biology
Bartosz Prokop, Lendert Gelens · 2025
Read Data-driven discovery of dynamical models in biology by Bartosz Prokop, Lendert Gelens (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-06735.txt.
Introduction
Learning the equations of a living system straight from data is reviewed by Bartosz Prokop and one co-author in a 2025 paper on computing in biology. Dynamical systems theory describes how living parts change together, from tiny chemical clocks to tissue patterns. Curved feedbacks, delays, and many scales make hand-built models hard as readings grow richer.
Three groups of data-driven methods are weighed: regression-based methods, network-based designs, and splitting techniques. Each is judged on three aims: seeing what comes next, finding which parts sway each other, and describing steady states, swings, and shifts. To weigh them fairly, sample methods are run on one shared test case, the Oregonator, a minimal chemical oscillator with design principles common in biology. Strengths, limits, and clarity are weighed so research teams can pick tools fit for complex, many-sided living data.
Catalog record
| CREATOR | Bartosz Prokop, Lendert Gelens |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-06735
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Data-driven discovery of dynamical models in biology Preprint: APS/123-QED Bartosz Prokop, Lendert Gelens Abstract Dynamical systems theory describes how interacting quantities change over time and space, from molecular oscillators to large-scale biological patterns. Such systems often involve nonlinear feedbacks, delays, and interactions across scales. Classical modeling derives explicit governing equations, often systems of differential equations, by combining mechanistic assumptions, experimental observations, and known physical laws. The growing complexity of biological processes has, however, motivated complementary data-driven methods that aim to infer model structure directly from measurements, often without specifying equations a priori. In this review, we survey approaches for model discovery in biological dynamical systems, focusing on three methodological families: regression-based methods, network-based architectures, and decomposition techniques. We compare their ability to address three core goals: forecasting future states, identifying interactions, and characterizing system states. Representative methods are applied to a common benchmark, the Oregonator model, a minimal nonlinear oscillator that captures shared design principles of chemical and biological systems. By highlighting strengths, limitations, and interpretability, we aim to guide researchers in selecting tools for analyzing complex, nonlinear, and high-dimensional dynamics in the life sciences. I Introduction I.1 Dynamical systems across science The world around us is constantly changing. From planetary orbits that govern the change of seasons, to turbulent weather patterns, to the molecular choreography of living cells: all are examples of dynamical systems , in which the state of a system evolves over time according to underlying rules. A dynamical system describes how a set of interacting quantities changes in time, and is often represented by a system of ordinary differential equations (ODEs). An ODE specifies how the rate of change of a quantity depends on the current state of the system. In vector form, a general continuous-time dynamical system can be written as Strogatz 2000 : \frac{d}{dt}X(t)=X_{t}(t)=\mathrm{f}(X(t),\beta(t)), (1) where X is the state vector containing the variables x_{i} with i=1,\dots,N , \mathrm{f} is the vector field governing the dynamics, and \beta is the set of parameters that may depend on time t . In experimental or computational studies, we often work with discrete-time representations: X_{k+1}=\mathrm{F}(X_{k}), (2) where X_{k} is the system state at time t_{k} and X_{k+1} at t_{k+1} , updated by the map \mathrm{F} . As illustrated in Fig. 1 A, small parameter changes can profoundly alter the qualitative behavior of a system, for example shifting it from a stable fixed point to sustained oscillations through a bifurcation. Mathematical models of dynamical systems allow us to predict future states, identify interactions between variables, and understand the mechanisms producing observed behavior. This modeling framework has long been central to physics and engineering, and is now also essential in the life sciences. I.2 Dynamical systems in biology Biological systems display rich spatiotemporal dynamical behavior across scales Tyson et al. 2003 ; Goldbeter et al. 2012 ; Gelens et al. 2014 ; Beta and Kruse 2017 ; Deneke and Di Talia 2018 ; Volpert and Petrovskii 2009 ; Goldbeter 2018 . Such dynamics may be purely temporal, describing how system variables change over time at a fixed location, or they may involve spatial propagation and pattern formation. Purely temporal examples include the cell cycle oscillator that governs cell division Ferrell et al. 2011 ; Novak and Tyson 1993 , circadian clocks that synchronize physiology with the day–night cycle Dunlap 1999 ; Bell-Pedersen et al. 2005 , intracellular calcium oscillations that regulate fertilization and neuronal signaling Berridge and Galione 1988 ; Borghans et al. 1997 , glycolytic oscillations in yeast metabolism Ghosh and Chance 1964 ; Chandra et al. 2011 , p53 oscillations in the cellular DNA damage response Geva-Zatorsky et al. 2006 ; Batchelor et al. 2008 ; Purvis et al. 2012 , and excitability in neurons, where action potential pulses arise from threshold-triggered ionic currents Hodgkin and Huxley 1939 ; Hodgkin and Huxley 1952 ; Izhikevich 2007 . Spatiotemporal examples include mitotic waves in early embryonic development Chang and Ferrell Jr 2013 ; Nolet et al. 2020 ; Afanzar et al. 2020 ; Puls et al. 2024 ; Deneke et al. 2016 , cyclic AMP waves in social amoebae during aggregation Devreotes et al. 1983 ; Tyson and Murray 1989 , and traveling waves of Erk activity discovered in zebrafish bone regeneration, where they regulate tissue growth De Simone et al. 2021 . These systems often involve nonlinear feedbacks, time delays, threshold responses, and spatial coupling, which together generate complex patterns in time and space (see Fig. 1 B–E for concrete examples). Figure 1: From data to dynamical systems in chemistry and biology. A. General representation of a dynamical system. The state vector X(t) evolves according to governing function f and parameters \beta . Small parameter changes can lead to qualitatively different time series, summarized in a bifurcation diagram where a stable fixed point loses stability and gives rise to a limit cycle. B–C. The Belousov–Zhabotinsky (BZ) reaction: experimental recordings (Adapted form Field et al. 1972 ) and spatial patterns Morris 2009 (B), alongside the Oregonator model (C), which reduces the chemistry to three interacting intermediates ( [\mathrm{Br}^{-}] , [\mathrm{HBrO}_{2}] , [\mathrm{Ce}^{4+}] ). D–E. The embryonic cell cycle oscillator: experimental observations of Cdk1 dynamics (Adapted form Rombouts et al. 2025 ) and mitotic (spiral) waves Cebrián-Lacasa et al. 2024 (D), together with a reduced three-variable model Parra-Rivas et al. 2023 (E) capturing the interactions among cyclin B, active cyclin B - Cdk1 (Cdk1 a ), and APC/C. F. From data to understanding. Ideally, one recovers the full governing equations f(X(t),\beta) , enabling forecasting , reconstruction of interaction networks , and identification of dynamic states . Alternatively, data-driven approaches can aim directly at these three complementary outcomes. I.3 Two illustrative oscillators in chemistry and biology To illustrate how dynamical systems arise in both chemistry and biology, we examine two illustrative oscillators. The first occurs in the Belousov-Zhabotinsky (BZ) reaction , initially described by Belousov in the 1950s Winfree 1984 and belonging to the family of (bio)chemical systems arising from non-equilibrium thermodynamics. Belousov observed periodic color changes in a mixture of bromate, cerium, and malonic acid, a phenomenon initially dismissed by the scientific community for apparently contradicting established chemical principles Winfree 1984 . Zhabotinsky later reproduced the reaction, and Field, Körös, and Noyes elucidated its underlying mechanism Field et al. 1972 . Experiments by these authors are shown in Fig. 1 B, alongside a Petri dish demonstration of the classical BZ reaction producing temporally changing wave patterns. Over the years, numerous mathematical models of the BZ reaction have been proposed, ranging from detailed mechanistic descriptions involving many intermediate species to reduced descriptions capturing only the dominant dynamical features Gyorgyi et al. 1990 ; Tyson 2013 ; Tyson 1977 ; Epstein and Pojman 1998 . Here, we focus on the Oregonator model developed by Field and Noyes Field and Noyes 1974 and written in a normalized form by Tyson Tyson 1985 (Fig. 1 C), a minimal set of nonlinear ODEs that retains only the three key intermediates and their essential interactions. We present this model because it captures the core feedback structure responsible for oscillations while remaining analytically and computationally tractable. In compact form: \begin{split}u_{t}&=p[u(1-u)-v(u-q)],\\ v_{t}&=p^{\prime}(uv+gv-2fw),\\ w_{t}&=u-w,\end{split} (3) where u=[\text{HBrO}_{2}] , v=[\text{Br}^{-}] , and w=[\text{Ce}^{4+}] represent the concentrations of the key