Embracing multimodal optimization to enhance Dynamic Energy Budget parameterization
Introduction
Dynamic Energy Budget (DEB) theory (Kooijman 2010) mechanistically explains individual bioenergetics throughout the life cycle under dynamic environmental conditions. Building on thermodynamics first-principles and assuming that the mechanisms responsible for the organization of individual metabolism are not species-specific (Sousa et al. 2008), DEB theory can be applied to all species. The potential of such a mechanistic theory has led to an increase in its popularity in the scientific literature, being applied to more than 2,000 species (Add my Pet, https://www.bio.vu.nl/thb/deb/deblab/add_my_pet/), among which bivalves are one of the most studied group (e.g. Pouvreau et al. 2006, Rosland et al. 2009). Despite the successful application of the theory, DEB has been criticized for not being ‘efficient’ due to the large number of parameters, namely 14 in its standard versions, and the fact that most of them are species-specific (Marquet et al. 2014). The complex and data-demanding procedure to estimate the value of these parameters (Lika et al., 2011a) may also limit its expansion. Contrarily, it has been argued that DEB is efficient given the large number of processes that can be derived from those parameters, for instance, life cycle development, feeding, growth, maintenance, metabolic heating, reproduction, and senescence (Kearney et al. 2015). The large suite of predictions that can be derived is precisely one of the strengths of DEB implementation in ecosystem-scale models (e.g. Guyondet et al. 2010).
The need to parameterize such a relatively large number of parameters, most of them species-specific, is one of the most challenging aspects of DEB modelling. Early on, the parameterization of DEB relied on the empirical and independent approximation to each parameter (e.g. van der Veer et al. 2006). Mathematical calibration such as the use of the Nelder-Mead method was also implemented to estimate the values of unknown parameters or to improve model fit in populations of the same species (e.g. Bacher and Gangnery 2006). A breakthrough in DEB parameterization came with the introduction of the “covariation method” (Lika et al. 2011a), a non-linear least squares regression approach. Lika et al. (2011a) suggested that the most robust parameterization would be to simultaneously estimate all DEB parameters by minimizing the discrepancies between simulations and all available observations. These available observations range from single data-points that represent a single value (zero-variate data in DEB jargon) such as maximum reproduction rate, to arrays of values (uni-variate data) such as growth over time. This method also takes into account the inter-dependency of some parameters, avoiding potentially incoherent combinations of parameters. In addition, Lika et al. (2011a) used DEB theory principles to constraint the potential range of solutions, and introduced the concept of pseudo-data, which are well-known values for a generalized animal that could be scaled to the modelled species. The use of pseudo-data has demonstrated to be beneficial to exploit existing knowledge about parameter values that cannot be extracted from observations, and it constitutes a key component to confine potential solutions to those that make sense biologically (Marques et al. 2019). Marques et al. (2019) has replaced Lika et al. (2011a) as the standard procedure to estimate DEB parameters by adding more advanced loss functions and filters to prevent parameter sets that are not compatible with the general principles of the theory.
Despite this progress, there are still major challenges in DEB parameter estimation, which are common to most mechanistic models in biology. For example, interindividual variability is included through population-level averages, but not directly included in the parameter estimation, and similarly, the uncertainty in the estimated parameters is not usually considered (Johnson et al. 2013). Some mathematical approaches have already included this uncertainty by applying a state-space method to determine the probability distribution of the estimated parameters (Fujiwara et al. 2005). Similarly, the use of a Bayesian inference framework allows for the consideration of this uncertainty (Johnson et al. 2013), and it has already been used to estimate DEB parameters (Boersch-Supan and Johnson 2018). Another challenge for a robust estimation in DEB is related to the large number of parameters that could lead to ‘the curse of dimensionality’ type of problems (Jusup et al. 2017). The challenge of dimensionality is that an increase in the number of parameters requires an exponential increase in available empirical data to avoid sparsity and obtain parameters that are statistically significant. This problem is magnified by the fact that some DEB parameters covariate (Fujiwara et al. 2005, Chica et al. 2017), suggesting that different combinations of values can result in the same solution; which could lead to the simulation of the right solutions for the wrong reasons. For example, the same growth could be observed with high feeding and high metabolic costs or with low feeding and low costs. Therefore, the advantage of DEB regarding the potential to simulate a large variety of processes could become a weakness if the available datasets do not allow for a meaningful parameterization that addresses this multimodal optimization problem, understood as a problem without a unique global optimum solution but multiple optima, either global or local (Ehrgott 2005). The multidimensional space of DEB parameters together with data-gaps could be considered one of the major hurdles for DEB parameterization.
Traditionally, the uni-variate data used for DEB calibration focus on growth curves, biometric data, and datasets that can inform about the effect of temperature on the physiology of the individual. Among these datasets, long-term growth curves are usually preferred given the relative simplicity to collect them, their availability in the literature, and the straightforward calculation of growth using DEB outputs; but they lack the power to inform about short-term individual responses. Recent optimization algorithms consider the coexistence of model parameters that are mapped to a multimodal fitness landscape (e.g. MOMCA framework, Chica et al. 2017). They aim to simultaneously locate more than one optimal solution. Subsequently, such approaches represent the nature of DEB theory and they could thus be an ideal way to move forward on DEB parameterization by providing alternative solutions rather than a single one. In this study, a combined dataset (Strohmeier et al. 2015) of growth and physiological rates (respiration and clearance rates) of the mussel Mytilus edulis from two locations of the Lysefjord (Norway) was used to parameterize a DEB model using a novel optimization approach that considers that DEB parameters are mapped to a multimodal fitness landscape (Chica et al. 2017). The inclusion of physiological rates in the parameterization aims to constrain the range of parameters and inform about short-term physiological responses. The ultimate goal of this study is not to provide a generalized set of parameters for M. edulis, but to demonstrate that this novel optimization can provide alternative solutions that could identify knowledge gaps and research priorities for further improvement of DEB parameterization. Furthermore, the provision of alternative equally good solutions becomes an ideal complement to the standard procedures to estimate DEB parameters (Marques et al. 2019).
Section snippets
Dynamic Energy Budget model
DEB theory describes the energy of an individual in terms of three state variables: reserve(s), structure(s), and maturity/reproduction. In brief, the assimilated energy is stored as reserves; a fixed fraction of the mobilized energy (κ) is then directed towards maintenance and growth of the structural body and the remainder (1-κ) is directed towards maturity maintenance and maturation or gamete production, depending on the life cycle stage of the organism (Fig. 1). A description of model
Global calibration
The best MOMCA set of parameters for the simultaneous global optimization of the four datasets used for calibration (Table 2) averaged an ERR of 0.115 (Table 3), with a total of 3 and 15 solutions within a range from the best solution of 0.001 and 0.002, respectively. The fitness diverged among the different types of measurements: shell length was the best predicted observation and clearance rate the worst, with an average ERR of 0.007 and 0.390, respectively (Table 3). As expected, the fitness
Discussion
Selecting the correct parameters of a model is key for its successful application. The multidimensional space in which DEB parameters coexist increases the complexity of parameterization. Furthermore, different parametrizations are typically mapped to equally optimal solutions (i.e., a multimodal landscape). In this study, the application of a novel mathematical tool together with the combination of growth and physiological data was used to estimate new sets of parameters for the Mytilus edulis
Conclusions
Individual based models such as DEB are critical tools to explore the current understanding of bivalve bioenergetics. These models are also cornerstone in ecosystem modelling, for example, in models that explore the effects of aquaculture on the environment or the suitability of an area for aquaculture development. However, the parameterization of these models is challenging. In this study, the MOMCA framework, a novel multimodal optimization approach, has been used to parameterize DEB for
Declaration of Competing Interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We would like to thank two anonymous reviewers who provided a thorough and constructive critique of our work. This work was funded by the Research Council of Norway (Project No. 196560), the Institute of Marine Research (Project No 14898), EXASOCO (PGC2018-101216-B-I00), and NSERC Discovery Grant to RF. MC acknowledges the Ramón y Cajal program (RYC-2016-19800).
References (46)
- et al.
Attachment strength of the mussel Mytilus galloprovincialis: effect of habitat and body size
Journal of Experimental Marine Biology and Ecology
(2013) - et al.
Use of dynamic energy budget and individual based models to simulate the dynamics of cultivated oyster populations
Journal of Sea Research
(2006) - et al.
Feeding and digestion by the mussel Mytilus edulis L. (Bivalvia: Mollusca) in mixtures of silt and algal cells at low concentrations
Journal of Experimental Marine Biology and Ecology
(1987) - et al.
Modelling growth and reproduction of the Pacific oyster Crassostrea gigas: advances in the oyster-DEB model through application to a coastal pond
Journal of Sea Research
(2009) - et al.
An individual-based population dynamic model for estimating biomass yield and nutrient fluxes through an off-shore mussel (Mytilus galloprovincialis) farm
Estuarine, Coastal and Shelf Science
(2009) - et al.
Multimodal optimization: an effective framework for model calibration
Information Sciences
(2017) - et al.
Mussels or tunicates: That is the question. Evaluating efficient and sustainable resource use by low-trophic species in aquaculture settings
Journal of Cleaner Production
(2019) - et al.
Modelling mussel shell and flesh growth using a dynamic net production approach
Aquaculture
(2019) - et al.
New insights into the reproductive cycle of two Great Scallop populations in Brittany (France) using a DEB modelling approach
Journal of Sea Research
(2019) - et al.
Integrating multiple spatial scales in the carrying capacity assessment of a coastal ecosystem for bivalve aquaculture
Journal of Sea Research
(2010)
Growth trajectories of shell and soft tissue in bivalves: seasonal variation in Mytilus edulis L
Journal of Experimental Marine Biology and Ecology
Gill dimensions, water pumping rate and body size in the mussel Mytilus edulis L
Journal of Experimental Marine Biology and Ecology
Physics of metabolic organization
Physics of Life Reviews
Fitting multiple models to multiple data sets
Journal of Sea Research
Algorithm selection for black-box continuous optimization problems: A survey on methods and challenges
Information Sciences
The “covariation method” for estimating the parameters of the standard Dynamic Energy Budget model I: Philosophy and approach
Journal of Sea Research
The “covariation method” for estimating the parameters of the standard Dynamic Energy Budget model II: Properties and preliminary patterns
Journal of Sea Research
Application of a dynamic energy budget model to the Pacific oyster, Crassostrea gigas, reared under various environmental conditions
Journal of Sea Research
Applying Dynamic Energy Budget (DEB) theory to simulate growth and bio-energetics of blue mussels under low seston conditions
Journal of Sea Research
A comprehensive analytical approach for policy analysis of system dynamics models
European Journal of Operational Research
DEB parameters estimation for Mytilus edulis
Journal of Sea Research
Modelling feeding processes in bivalves: a mechanistic approach
Ecological Modelling
The estimation of DEB parameters for various Northeast Atlantic bivalve species
Journal of Sea Research
Cited by (5)
An adaptive dynamic programming in cooperative target tracking for energy acquisition in wireless sensor networks
2023, Alexandria Engineering JournalApplied fish bioenergetics
2022, Fish PhysiologyCitation Excerpt :One of the key components of DEB modeling is the parameterization. Although some of the DEB parameters can be estimated using specific experiments (van der Veer et al., 2006), the most current approach is to use mathematical estimation (Filgueira et al., 2020; Lika et al., 2011). The rationale behind mathematical calibration is that a collection of all available data for the species and simultaneous estimation of all DEB parameters would provide a robust parameterization that would minimize the impact of uncertainty that any specific experiment could generate (Lika et al., 2011).
The role of Dynamic Energy Budgets in conservation physiology
2021, Conservation Physiology