Abstract
We propose a machine learning framework for the data-driven discovery of macroscopic chemotactic Partial Differential Equations (PDEs)—and the closures that lead to them- from high-fidelity, individual-based stochastic simulations of Escherichia coli bacterial motility. The fine scale, chemomechanical, hybrid (continuum—Monte Carlo) simulation model embodies the underlying biophysics, and its parameters are informed from experimental observations of individual cells. Using a parsimonious set of collective observables, we learn effective, coarse-grained “Keller–Segel class” chemotactic PDEs using machine learning regressors: (a) (shallow) feedforward neural networks and (b) Gaussian Processes. The learned laws can be black-box (when no prior knowledge about the PDE law structure is assumed) or gray-box when parts of the equation (e.g. the pure diffusion part) is known and “hardwired” in the regression process. More importantly, we discuss data-driven corrections (both additive and functional), to analytically known, approximate closures.
Similar content being viewed by others
Data availibility
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
Notes
We note here that the identification is performed in Euclidean space; in the case of spherical or cylindrical geometries we may need to express the right-hand-side not in terms of derivatives wrt. the independent variables, but rather in their coordinate-invariant form Psarellis et al. (2022).
References
Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z, Citro C, Corrado GS, Davis A, Dean J, Devin M, Ghemawat S, Goodfellow I, Harp A, Irving G, Isard M, Jia Y, Jozefowicz R, Kaiser L, Kudlur M, Levenberg J, Mané D, Monga R, Moore S, Murray D, Olah C, Schuster M, Shlens J, Steiner B, Sutskever I, Talwar K, Tucker P, Vanhoucke V, Vasudevan V, Viégas F, Vinyals O, Warden P, Wattenberg M, Wicke M, Yu Y, Zheng X (2015) TensorFlow: large-scale machine learning on heterogeneous systems, software available from tensorflow.org. https://www.tensorflow.org/
Adler J (1969) Chemoreceptors in bacteria. Science 166(3913):1588–1597
Alexandridis A, Siettos C, Sarimveis H, Boudouvis A, Bafas G (2002) Modelling of nonlinear process dynamics using Kohonen’s neural networks, fuzzy systems and Chebyshev series. Comput Chem Eng 26(4–5):479–486
Ansumali S, Frouzakis CE, Karlin IV, Kevrekidis IG (2005) Exploring Hydrodynamic Closures for the Lid-driven Micro-cavity. arXiv: Statistical Mechanics
Arbabi H, Kevrekidis IG (2021) Particles to partial differential equations parsimoniously. Chaos Interdiscip J Nonlinear Sci 31(3):033137
Beck A, Flad D, Munz C-D (2019) Deep neural networks for data-driven LES closure models. J Comput Phys 398:108910
Bellomo N, Bellouquid A, Nieto J, Soler J (2010) Multiscale biological tissue models and flux-limited chemotaxis for multicellular growing systems. Math Models Methods Appl Sci 20(07):1179–1207
Bellomo N, Outada N, Soler J, Tao Y, Winkler M (2022) Chemotaxis and cross-diffusion models in complex environments: models and analytic problems toward a multiscale vision. Math Models Methods Appl Sci 1–80
Berg HC, Brown DA (1972) Chemotaxis in Escherichia coli analysed by three-dimensional tracking. Nature 239(5374):500–504
Berg HC, Turner L (1990) Chemotaxis of bacteria in glass capillary arrays, Escherichia coli, motility, microchannel plate, and light scattering. Biophys J 58(4):919–930
Bertalan T, Dietrich F, Mezić I, Kevrekidis IG (2019) On learning Hamiltonian systems from data. Chaos Interdiscip J Nonlinear Sci 29(12):121107
Block SM, Segall JE, Berg HC (1982) Impulse responses in bacterial chemotaxis. Cell 31(1):215–226
Block SM, Segall JE, Berg HC (1983) Adaptation kinetics in bacterial chemotaxis. J Bacteriol 154(1):312–323
Bowman AW, Azzalini A (1997) Applied smoothing techniques for data analysis: the kernel approach with S-Plus illustrations, vol 18. OUP, Oxford
Boyd A, Krikos A, Simon M (1981) Sensory transducers of E. coli are encoded by homologous genes. Cell 26(3):333–343
Brunton SL, Proctor JL, Kutz JN (2016) Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proc Natl Acad Sci 113(15):3932–3937
Chavanis P-H (2008) Nonlinear mean field Fokker–Planck equations. Application to the chemotaxis of biological populations. Eur Phys J B 62(2):179–208
Chen RTQ, Rubanova Y, Bettencourt J, Duvenaud D (2019) Neural ordinary differential equations. arXiv:1806.07366
Chen T, Chen H (1995) Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems. IEEE Trans Neural Netw 6(4):911–917
Chen Y, Hosseini B, Owhadi H, Stuart AM (2021) Solving and learning nonlinear PDEs with Gaussian processes. J Comput Phys 447:110668
Chen Z, Churchill V, Wu K, Xiu D (2022) Deep neural network modeling of unknown partial differential equations in nodal space. J Comput Phys 449:110782
Cluzel P, Surette M, Leibler S (2000) An ultrasensitive bacterial motor revealed by monitoring signaling proteins in single cells. Science 287(5458):1652–1655
Coburn L, Cerone L, Torney C, Couzin ID, Neufeld Z (2013) Tactile interactions lead to coherent motion and enhanced chemotaxis of migrating cells. Phys Biol 10(4):046002
Dormand J, Prince P (1980) A family of embedded Runge–Kutta formulae. J Comput Appl Math 6(1):19–26
Dsilva CJ, Talmon R, Coifman RR, Kevrekidis IG (2018) Parsimonious representation of nonlinear dynamical systems through manifold learning: a chemotaxis case study. Appl Comput Harmon Anal 44(3):759–773. https://doi.org/10.1016/j.acha.2015.06.008
Duraisamy K, Iaccarino G, Xiao H (2019) Turbulence modeling in the age of data. Annu Rev Fluid Mech 51(1):357–377
Emonet T, Macal CM, North MJ, Wickersham CE, Cluzel P (2005) Agentcell: a digital single-cell assay for bacterial chemotaxis. Bioinformatics 21(11):2714–2721
Erban R, Othmer HG (2004) From individual to collective behavior in bacterial chemotaxis. SIAM J Appl Math 65(2):361–391
Erban R, Othmer HG (2007) Taxis equations for amoeboid cells. J Math Biol 54(6):847–885. https://doi.org/10.1007/s00285-007-0070-1
Erban R, Kevrekidis IG, Othmer HG (2006) An equation-free computational approach for extracting population-level behavior from individual-based models of biological dispersal. Physica D 215(1):1–24
Erban R, Frewen TA, Wang X, Elston TC, Coifman R, Nadler B, Kevrekidis IG (2007) Variable-free exploration of stochastic models: a gene regulatory network example. J Chem Phys 126(15):04B618
Franz B, Erban R (2013) Hybrid modelling of individual movement and collective behaviour. In: Dispersal, individual movement and spatial ecology. Springer, pp 129–157
Galaris E, Fabiani G, Gallos I, Kevrekidis I, Siettos C (2022) Numerical bifurcation analysis of PDEs from lattice Boltzmann model simulations: a parsimonious machine learning approach. J Sci Comput 92(2):34
Gonzalez-Garcia R, Rico-Martinez R, Kevrekidis I (1998) Identification of distributed parameter systems: a neural net based approach. Comput Chem Eng 22:S965–S968
Gorban AN, Kevrekidis IG, Theodoropoulos C, Kazantzis NK, Öttinger HC (Eds.) (2006) Model reduction and coarse-graining approaches for multiscale phenomena. Springer, Berlin https://doi.org/10.1007/3-540-35888-9
Heit B, Tavener S, Raharjo E, Kubes P (2002) An intracellular signaling hierarchy determines direction of migration in opposing chemotactic gradients. J Cell Biol 159(1):91–102
Ho KKY, Srivastava S, Kinnunen PC, Garikipati K, Luker GD, Luker KE (2023) Oscillatory ERK signaling and morphology determine heterogeneity of breast cancer cell chemotaxis via MEK-ERK and p38-MAPK signaling pathways. Bioengineering 10(2). https://doi.org/10.3390/bioengineering10020269
Ishihara A, Segall JE, Block SM, Berg HC (1983) Coordination of flagella on filamentous cells of Escherichia coli. J Bacteriol 155(1):228–237
Iskhakov AS, Dinh NT, Chen E (2021) Integration of neural networks with numerical solution of PDEs for closure models development. Phys Lett A 406:127456
Jiang Y, Kolehmainen J, Gu Y, Kevrekidis YG, Ozel A, Sundaresan S (2019) Neural-network-based filtered drag model for gas-particle flows. Powder Technol 346:403–413
Kamath A, Vargas-Hernández RA, Krems RV, Carrington T, Manzhos S (2018) Neural networks vs Gaussian process regression for representing potential energy surfaces: a comparative study of fit quality and vibrational spectrum accuracy. J Chem Phys 148(24):241702. https://doi.org/10.1063/1.5003074
Karniadakis GE, Kevrekidis IG, Lu L, Perdikaris P, Wang S, Yang L (2021) Physics-informed machine learning. Nat Rev Phys 3(6):422–440
Keller EF, Segel LA (1971) Model for chemotaxis. J Theor Biol 30(2):225–234
Kemeth FP, Bertalan T, Thiem T, Dietrich F, Moon SJ, Laing CR, Kevrekidis IG (2022) Learning emergent partial differential equations in a learned emergent space. Nat Commun 13(1):3318
Kim I, Yao Y (2012) The Patlak–Keller–Segel model and its variations: properties of solutions via maximum principle. SIAM J Math Anal 44(2):568–602
Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. https://doi.org/10.48550/ARXIV.1412.6980
Kocijan J, Girard A, Banko B, Murray-Smith R (2005) Dynamic systems identification with Gaussian processes. Math Comput Model Dyn Syst 11(4):411–424
Krischer K, Rico-Martinez R, Kevrekidis I, Rotermund H, Ertl G, Hudson J (1993) Model identification of a spatiotemporally varying catalytic reaction. AIChE J 39(1):89–98
Larsen SH, Reader RW, Kort EN, Tso W-W, Adler J (1974) Change in direction of flagellar rotation is the basis of the chemotactic response in Escherichia coli. Nature 249(5452):74–77
LeCun Y, Bengio Y (1998) Convolutional Networks for Images, Speech, and Time Series. MIT Press, Cambridge, pp 255–258
Lee S, Dietrich F, Karniadakis GE, Kevrekidis IG (2019) Linking Gaussian process regression with data-driven manifold embeddings for nonlinear data fusion. Interface Focus 9(3):20180083
Lee S, Kooshkbaghi M, Spiliotis K, Siettos CI, Kevrekidis IG (2020) Coarse-scale PDEs from fine-scale observations via machine learning. Chaos Interdiscip J Nonlinear Sci 30(1):013141
Lee K, Hernández AM, Stewart DS, Lee S (2021) Data-driven blended equations of state for condensed-phase explosives. Combust Theory Modell 1–23
Li J, Kevrekidis PG, Gear CW, Kevrekidis IG (2003) Deciding the nature of the coarse equation through microscopic simulations: the baby-bathwater scheme. Multiscale Model Simul 1(3):391–407
Liu J, Parkinson JS (1989) Role of chew protein in coupling membrane receptors to the intracellular signaling system of bacterial chemotaxis. Proc Natl Acad Sci 86(22):8703–8707
Liu K, Li Y, Hu X, Lucu M, Widanage WD (2019) Gaussian process regression with automatic relevance determination kernel for calendar aging prediction of lithium-ion batteries. IEEE Trans Industr Inf 16(6):3767–3777
MacKay DJ (1992) Bayesian interpolation. Neural Comput 4(3):415–447
Maeda K, Imae Y, Shioi J-I, Oosawa F (1976) Effect of temperature on motility and chemotaxis of Escherichia coli. J Bacteriol 127(3):1039–1046
Masri SF, Chassiakos AG, Caughey TK (1993) Identification of nonlinear dynamic systems using neural networks. J Appl Mech 60(1):123–133
Nash J (1966) Analyticity of the solutions of implicit function problems with analytic data. Ann Math 84(3):345–355
Othmer HG, Schaap P (1998) Oscillatory camp signaling in the development of Dictyostelium discoideum. Comments Theor Biol 5:175–282
Othmer HG, Xin X, Xue C (2013) Excitation and adaptation in bacteria-a model signal transduction system that controls taxis and spatial pattern formation. Int J Mol Sci 14(5):9205–9248
Painter KJ (2019) Mathematical models for chemotaxis and their applications in self-organisation phenomena. J Theor Biol 481:162–182
Pan S, Duraisamy K (2018) Data-driven discovery of closure models. SIAM J Appl Dyn Syst 17(4):2381–2413
Parish EJ, Duraisamy K (2016) A paradigm for data-driven predictive modeling using field inversion and machine learning. J Comput Phys 305:758–774
Parkinson JS (1976) cheA, cheB, and cheC genes of Escherichia coli and their role in chemotaxis. J Bacteriol 126(2):758–770
Parkinson JS (1980) Novel mutations affecting a signaling component for chemotaxis of Escherichia coli. J Bacteriol 142(3):953–961
Pathak J, Mustafa M, Kashinath K, Motheau E, Kurth T, Day M (2020) Using machine learning to augment coarse-grid computational fluid dynamics simulations. https://doi.org/10.48550/ARXIV.2010.00072
Patlak CS (1953) A mathematical contribution to the study of orientation of organisms. Bull Math Biophys 15(4):431–476
Perdikaris P, Raissi M, Damianou A, Lawrence ND, Karniadakis GE (2017) Nonlinear information fusion algorithms for data-efficient multi-fidelity modelling. Proc R Soc A Math Phys Eng Sci 473(2198):20160751
Psarellis YM, Lee S, Bhattacharjee T, Datta SS, Bello-Rivas JM, Kevrekidis IG (2022) Data-driven discovery of chemotactic migration of bacteria via machine learning. https://doi.org/10.48550/ARXIV.2208.11853
Qin T, Wu K, Xiu D (2019) Data driven governing equations approximation using deep neural networks. J Comput Phys 395:620–635
Raissi M, Karniadakis GE (2018) Hidden physics models: machine learning of nonlinear partial differential equations. J Comput Phys 357:125–141
Raissi M, Perdikaris P, Karniadakis GE (2017) Machine learning of linear differential equations using Gaussian processes. J Comput Phys 348:683–693
Raissi M, Perdikaris P, Karniadakis GE (2019) Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys 378:686–707
Rao C, Ren P, Liu Y, Sun H (2022) Discovering nonlinear PDEs from scarce data with physics-encoded learning. arXiv:2201.12354
Rasmussen CE, Williams CKI (2005) Gaussian processes for machine learning (adaptive computation and machine learning). The MIT Press
Rasmussen CE, Williams CKI (2006) Gaussian processes for machine learning. MIT Press
Rico-Martinez R, Krischer K, Kevrekidis I, Kube M, Hudson J (1992) Discrete-vs. continuous-time nonlinear signal processing of cu electrodissolution data. Chem Eng Commun 118(1):25–48
Rico-Martinez R, Anderson J, Kevrekidis I (1994) Continuous-time nonlinear signal processing: a neural network based approach for gray box identification. In: Proceedings of IEEE workshop on neural networks for signal processing. IEEE, pp 596–605
Rousset M, Samaey G (2013) Simulating individual-based models of bacterial chemotaxis with asymptotic variance reduction. Math Models Methods Appl Sci 23(12):2155–2191
Sandhu R, Pettit C, Khalil M, Poirel D, Sarkar A (2017) Bayesian model selection using automatic relevance determination for nonlinear dynamical systems. Comput Methods Appl Mech Eng 320:237–260
Sarkar MK, Paul K, Blair D (2010) Chemotaxis signaling protein CheY binds to the rotor protein FliN to control the direction of flagellar rotation in Escherichia coli. Proc Natl Acad Sci 107(20):9370–9375
Scharf BE, Fahrner KA, Turner L, Berg HC (1998) Control of direction of flagellar rotation in bacterial chemotaxis. Proc Natl Acad Sci 95(1):201–206
Segel LA, Goldbeter A, Devreotes PN, Knox BE (1986) A mechanism for exact sensory adaptation based on receptor modification. J Theor Biol 120(2):151–179
Setayeshgar S, Gear CW, Othmer HG, Kevrekidis IG (2005) Application of coarse integration to bacterial chemotaxis. Multiscale Model Simul 4(1):307–327
Sheriffdeen S, Ragusa JC, Morel JE, Adams ML, Bui-Thanh T (2019) Accelerating PDE-constrained inverse solutions with deep learning and reduced order models. arXiv:1912.08864
Siettos C (2014) Coarse-grained computational stability analysis and acceleration of the collective dynamics of a Monte Carlo simulation of bacterial locomotion. Appl Math Comput 232:836–847
Siettos CI, Bafas GV (2002) Semiglobal stabilization of nonlinear systems using fuzzy control and singular perturbation methods. Fuzzy Sets Syst 129(3):275–294
Siettos CI, Bafas GV, Boudouvis AG (2002) Truncated Chebyshev series approximation of fuzzy systems for control and nonlinear system identification. Fuzzy Sets Syst 126(1):89–104
Spiro PA, Parkinson JS, Othmer HG (1997) A model of excitation and adaptation in bacterial chemotaxis. Proc Natl Acad Sci 94(14):7263–7268
Takens F (1981) Detecting strange attractors in turbulence. In: Rand D, Young L-S (eds) Dynamical Systems and Turbulence, Warwick 1980. Springer, Berlin, pp 366–381
Thiem TN, Kemeth FP, Bertalan T, Laing CR, Kevrekidis IG (2021) Global and local reduced models for interacting, heterogeneous agents. Chaos Interdiscip J Nonlinear Sci 31(7):073139
Tindall MJ, Porter S, Maini P, Gaglia G, Armitage JP (2008) Overview of mathematical approaches used to model bacterial chemotaxis I: the single cell. Bull Math Biol 70(6):1525–1569
Tindall MJ, Maini PK, Porter SL, Armitage JP (2008) Overview of mathematical approaches used to model bacterial chemotaxis II: bacterial populations. Bull Math Biol 70(6):1570
Turner L, Caplan SR, Berg HC (1996) Temperature-induced switching of the bacterial flagellar motor. Biophys J 71(4):2227–2233
Vlachas PR, Byeon W, Wan ZY, Sapsis TP, Koumoutsakos P (2018) Data-driven forecasting of high-dimensional chaotic systems with long short-term memory networks. Proc R Soc A Math Phys Eng Sci 474(2213):20170844
Vlachas PR, Pathak J, Hunt BR, Sapsis TP, Girvan M, Ott E, Koumoutsakos P (2020) Backpropagation algorithms and reservoir computing in recurrent neural networks for the forecasting of complex spatiotemporal dynamics. Neural Netw 126:191–217
Vlachas PR, Arampatzis G, Uhler C, Koumoutsakos P (2022) Multiscale simulations of complex systems by learning their effective dynamics. Nat Mach Intell 4(4):359–366
Wan ZY, Sapsis TP (2017) Reduced-space Gaussian process regression for data-driven probabilistic forecast of chaotic dynamical systems. Physica D 345:40–55
Whitney H (1936) Differentiable manifolds. Ann Math 37(3):645–680
Wu K, Xiu D (2019) Numerical aspects for approximating governing equations using data. J Comput Phys 384:200–221
Wu K, Xiu D (2020) Data-driven deep learning of partial differential equations in modal space. J Comput Phys 408:109307
Wu M, Roberts JW, Kim S, Koch DL, DeLisa MP (2006) Collective bacterial dynamics revealed using a three-dimensional population-scale defocused particle tracking technique. Appl Environ Microbiol 72(7):4987–4994
Xue C (2015) Macroscopic equations for bacterial chemotaxis: integration of detailed biochemistry of cell signaling. J Math Biol 70(1):1–44
Yang L, Meng X, Karniadakis GE (2021) B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data. J Comput Phys 425:109913
Yasuda S (2017) Monte Carlo simulation for kinetic chemotaxis model: an application to the traveling population wave. J Comput Phys 330:1022–1042
Zhang ZJ, Duraisamy K (2015) ‘Machine learning methods for data-driven turbulence modeling’. 22nd AIAA computational fluid dynamics conference, american institute of aeronautics and astronautics. https://doi.org/10.2514/6.2015-2460. AIAA AVIATION Forum
Acknowledgements
This work was partially supported by the US Department of Energy, by the US Air Force Office of Scientific Research and by DARPA. C. S. was partially supported by INdAM, through GNCS and the Italian research fund FISR2020IP - 02893.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors have no competing interests to declare that are relevant to the content of this article.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Details on the Monte Carlo chemotaxis model
Our microscopic, agent-based model is based on the work of Othmer and Schaap (1998); Othmer et al. (2013). Each bacterium is modelled as having six flagellae; special care has been taken in modelling the direction of the rotation of the flagellar filaments, as this constitutes the basis of chemotaxis (Larsen et al. 1974; Spiro et al. 1997). Following Scharf et al. (1998), the motor dynamics are described by a two-state system modelling the transition rates (transition probabilities per unit time) between CCW and CW (counter-clockwise and clockwise, respectively) rotation for each flagellum. These are characterized by an exponential distribution of time intervals in each state (Turner et al. 1996). Let us denote by \(k^+\) (\(k^-\)) the transition rate from CCW to CW (CW to CCW). Then, the bias of CW, i.e. the fraction of time that a flagellum rotates CW is \(p_{CW}=\frac{k^+}{k^{+}+k^-}\), \(p_{CCW}=1-p_{CW}\). The reversal frequency in the direction of rotation of the flagellar motors is Turner et al. (1996):
and the rate constants are given by:
For a cell with N flagellae, the total CCW bias of the cell is given by Spiro et al. (1997):
For \(N=6\), \(p_{CCW}=0.64\) and \(\theta =N/2\), we get \(P_{CCW}\sim 0.87\) suggesting that the cell spends around 90% of the time running (Spiro et al. 1997). This result is in line with experimental observations for the motility of wild-type E. coli in the absence of changes of the substrate, where the mean run (swimming) periods are \(\sim 1\) s and the tumble periods \(\sim 0.1\) s (for the strain AW405 in dilute phosphate buffer at \(32^{\circ }\hbox {C}\)) (Ishihara et al. 1983).
Experimental studies have shown that these rates depend on the CheY-P concentration, say C. In particular, Cluzel et al. (2000) have shown that the dependence of CW bias (between the values 0.1 and 0.9) to C can be approximated by a Hill function with a coefficient \(H \sim 10.3 \pm 1.1\), with a dissociation constant \(K_d= 3.1\) mM/s. Thus, the CW bias reads:
Based on the above findings, the transition rates \(k^+\), \(k^-\) are given by Setayeshgar et al. (2005):
Thus, based on the model formulation and nominal values of the parameters, the expected fraction of time spent in the CCW state in the absence of stimulus for each cell from kMC simulations is \(\sim \) 0.855, close enough to the one observed experimentally. In the absence of spatial variations in the chemoattractant (or repellent) profile, the rotation of the flagellar filament is biased towards the CCW direction (that is, the probability of CCW rotation of a flagellum is higher than that of CW rotation), when viewed along the helix axis towards the point of insertion in the cell (Larsen et al. 1974). This bias depends on the type of bacterial strain and the temperature; for the wild-type strain AW405, it has been found that the average value of the CCW bias is 0.64 at \(32^{\circ }\hbox {C}\) (Larsen et al. 1974; Block et al. 1982). When the majority of the flagellar filaments rotate CCW (CW) the cell swims (tumbles).
Appendix B: Determination of the parameters of the macroscopic PDE for bacterial density evolution
1.1 Appendix B.1: Determination of the diffusion coefficient
An estimation of the diffusion coefficient for cell motility in the absence of stimulus can be attempted following two paths. From a microscopic point of view, considering a random walk simulation, the mean free path i.e., the swimming distance without any change in the direction is given by \(\delta r = \tau \cdot \bar{v}\), where \(\tau \) is the mean time of swimming in one direction. Considering n such time steps in time t (i.e. \(n=t/\tau \)), the total mean-squared displacement \(\Delta r (t)^2\) at a certain time (t) is given by the Einstein relation (Berg and Turner 1990):
which is valid for \(t>> \tau \), where, \(\tau \) is the characteristic time scale. Here, the value of \(D_m\) is estimated from our Monte Carlo simulations in the absence of stimulus (we have set \(s(x)=1\), \(\forall x\)), by tracking the trajectories of 1000 cells for a time period of \(2000~\textrm{s}\). The cells are initially positioned at the middle of the domain, all initialized at the tumbling phase, with \(u_1(0)=0\) (no excitation), and adapted with \(u_2(0)=f(s(x))\), with a constant velocity of \(\bar{v}=0.003~\hbox {cm}/\hbox {s}\) (as in Berg and Turner 1990). Figure 5 depicts the average of the square distance as a function of time. By least-squares, we get \(\hat{D}_m \approx 9 \cdot 10^{-6}~\hbox {cm}^{2}/\hbox {s}\). This is in good agreement with experimental observations for the E. coli motility (see Berg and Brown 1972; Berg and Turner 1990; Spiro et al. 1997; Cluzel et al. 2000).
From a macroscopic point of view, one can estimate the diffusion coefficient \(D_M\) from a linear curve fitting between \(\frac{\partial b}{\partial t}\) and \(\frac{\partial ^2 b}{\partial x^2}\) with finite difference approximations of temporal and spatial derivatives at the coarse-scale. Thus, by fixing a spatial gradient of chemo-nutrient profile to zero (\(\nabla c = 0\)), we can consider a simple diffusion equation with a constant diffusion coefficient, D:
Finally, we note that the Einstein relation for the diffusion coefficient given by Eq. (B.1) can be approximated on average over a run and tumble period as:
where \({\bar{T}_{run}, \bar{T}_{tumb}}\) denote the average duration of swimming and tumbling periods, respectively, and \(\bar{v}\) is the average swimming speed. Thus, based on Eq. (B.3) and assuming that the tumbling duration is negligible compared to the swimming duration (as assumed for the derivation of the generalized Keller–Segel theory embodied in Eq. (3)), an approximation of the diffusion coefficient is given by:
Hence, setting \(\bar{D}_m=\hat{D}_m\approx 9 \cdot 10^{-6}~\hbox {cm}^2/\hbox {s}\), \(\lambda _0=1~\hbox {s}^{-1}\) (in agreement with experimental findings), we get \(\bar{v}=\sqrt{2}v=3 \sqrt{2} \cdot 9 \cdot 10^{-3}~\hbox {cm}/\hbox {s}\) as the average velocity appearing in Eq. (3).
1.2 Appendix B.2: Determination of the parameter c of the macroscopic PDE
As stated in Sect. 2, one of the assumptions for the derivation of the closed-form Keller–Segel Eq. (3) is the linear relation between the turning frequency \(\lambda \), and the basal frequency \(\lambda _0 \sim 1~\hbox {s}^{-1}\), i.e. for each cell at position x at time t, we have (see Eq. (5)):
For initial values \(u_1(x,0)\), \(u_2(x,0)\) for all cells (i.e. \(\forall x \in \mathrm I\!R\)), the analytical solution of the cartoon model (Eq. 2) is given by:
Note that, if one sets as initial value \(u_2(x,0)=f(s)\), then the second equation of the cartoon model (see Eq. (2)) gives \(u_2(x,t)=f(s)\), \(\forall t\) and the analytical solution for \(u_1(x,t)\) is reduced to:
To this end, the parameter c in Eq. (5) appearing in the Keller–Segel-class PDE given by Eq. (3) can be found with the aid of Monte Carlo simulations, by fixing \(u_{1}(x,t)\), \(\forall t\) to different relatively small values, say \(u_1\) \(\forall x\), and measuring, the number of turning events \(\lambda (u_1)\); then the value of the parameter c can be estimated by least-squares. A different way would be to set an initial value for \(u_1(x,0)\) (setting also as initial value \(u_2(x,0)=f(s)\)), run the Monte Carlo simulator, measure the turning frequencies \(\lambda (u_1(t))\) and based on the above, the value of c can be again estimated with least-squares.
Here, to estimate c, we have fixed \(u_1\) to the following values: \(-0.02\), \(-0.015\), \(-0.01\), \(-0.005\), 0, 0.005, 0.01, 0.015, 0.02, where the linear relation between \(\lambda \) and \(\lambda _0\) is valid, and we computed \(\lambda (u_1)\) based on Monte Carlo simulations with 1000 cells for a time period of 2000s. For these values, \(\hat{\lambda _0}\sim 1 s^{-1}\) (in a good agreement with the experimental findings) and \(\hat{c}\sim 19.5\) (99% CI 18–21). We note that this value is consistent with what has been reported in other studies (Xue 2015). For our simulations with the macroscopic PDE, we have set \(\hat{c}=20\).
Appendix C: Numerical details
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Lee, S., Psarellis, Y.M., Siettos, C.I. et al. Learning black- and gray-box chemotactic PDEs/closures from agent based Monte Carlo simulation data. J. Math. Biol. 87, 15 (2023). https://doi.org/10.1007/s00285-023-01946-0
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00285-023-01946-0
Keywords
- Inverse problems
- Partial differential equations
- Machine learning
- Stochastic simulations
- Chemotaxis
- Numerical analysis
- Multiscale methods