Calibration and evaluation of individual-based models using Approximate Bayesian Computation

This paper investigates the feasibility of using approximate Bayesian computation (ABC) to calibrate and evaluate complex individual-based models (IBMs). As ABC evolves, various versions are emerging, but here we only explore the most accessible version, rejection-ABC. Rejection-ABC involves running models a large number of times, with parameters drawn randomly from their prior distributions, and then retaining the simulations closest to the observations. Although well-established in some fields, whether ABC will work with ecological IBMs is still uncertain. Rejection-ABC was applied to an existing 14-parameter earthworm energy budget IBM for which the available data consist of body mass growth and cocoon production in four experiments. ABC was able to narrow the posterior distributions of seven parameters, estimating credible intervals for each. ABC’s accepted values produced slightly better fits than literature values do. The accuracy of the analysis was assessed using cross-validation and coverage, currently the best available tests. Of the seven unnarrowed parameters, ABC revealed that three were correlated with other parameters, while the remaining four were found to be not estimable given the data available. It is often desirable to compare models to see whether all component modules are necessary. Here we used ABC model selection to compare the full model with a simplified version which removed the earthworm’s movement and much of the energy budget. We are able to show that inclusion of the energy budget is necessary for a good fit to the data. We show how our methodology can inform future modelling cycles, and briefly discuss how more advanced versions of ABC may be applicable to IBMs. We conclude that ABC has the potential to represent uncertainty in model structure, parameters and predictions, and to embed the often complex process of optimizing an IBM’s structure and parameters within an established statistical framework, thereby making the process more transparent and objective.

Predicting how many animals will be where: how to build, calibrate and evaluate individual-based models

Individual-based models (IBMs) can simulate the actions of individual animals as they interact with one another and the landscape in which they live. When used in spatially-explicit landscapes IBMs can show how populations change over time in response to management actions. For instance, IBMs are being used to design strategies of conservation and of the exploitation of fisheries, and for assessing the effects on populations of major construction projects and of novel agricultural chemicals. In such real world contexts, it becomes especially important to build IBMs in a principled fashion, and to approach calibration and evaluation systematically. We argue that insights from physiological and behavioural ecology offer a recipe for building realistic models, and that Approximate Bayesian Computation (ABC) is a promising technique for the calibration and evaluation of IBMs. IBMs are constructed primarily from knowledge about individuals. In ecological applications the relevant knowledge is found in physiological and behavioural ecology, and we approach these from an evolutionary perspective by taking into account how physiological and behavioural processes contribute to life histories, and how those life histories evolve. Evolutionary life history theory shows that, other things being equal, organisms should grow to sexual maturity as fast as possible, and then reproduce as fast as possible, while minimising per capita death rate. Physiological and behavioural ecology are largely built on these principles together with the laws of conservation of matter and energy. To complete construction of an IBM information is also needed on the effects of competitors, conspecifics and food scarcity; the maximum rates of ingestion, growth and reproduction, and life-history parameters. Using this knowledge about physiological and behavioural processes provides a principled way to build IBMs, but model parameters vary between species and are often difficult to measure. A common solution is to manually compare model outputs with observations from real landscapes and so to obtain parameters which produce acceptable fits of model to data. However, this procedure can be convoluted and lead to over-calibrated and thus inflexible models. Many formal statistical techniques are unsuitable for use with IBMs, but we argue that ABC offers a potential way forward. It can be used to calibrate and compare complex stochastic models and to assess the uncertainty in their predictions. We describe methods used to implement ABC in an accessible way and illustrate them with examples and discussion of recent studies. Although much progress has been made, theoretical issues remain, and some of these are outlined and discussed.

Transmission of Chlamydia trachomatis through sexual partnerships: a comparison between three individual-based models and empirical data

Chlamydia trachomatis is the most common bacterial sexually transmitted infection (STI) in many developed countries. The highest prevalence rates are found among young adults who have frequent partner change rates. Three published individual-based models have incorporated a detailed description of age-specific sexual behaviour in order to quantify the transmission of C. trachomatis in the population and to assess the impact of screening interventions. Owing to varying assumptions about sexual partnership formation and dissolution and the great uncertainty about critical parameters, such models show conflicting results about the impact of preventive interventions. Here, we perform a detailed evaluation of these models by comparing the partnership formation and dissolution dynamics with data from Natsal 2000, a population-based probability sample survey of sexual attitudes and lifestyles in Britain. The data also allow us to describe the dispersion of C. trachomatis infections as a function of sexual behaviour, using the Gini coefficient. We suggest that the Gini coefficient is a useful measure for calibrating infectious disease models that include risk structure and highlight the need to estimate this measure for other STIs.

A New Class of Processes for Formalizing and Generalizing Individual-Based Models: The Semi-Semi-Markov Processes

2000 Mathematics Subject Classification: 60K15, 60K20, 60G20,60J75, 60J80, 60J85, 60-08, 90B15.

Representing the acquisition and use of energy by individuals in agent-based models of animal populations

Summary 1. Agent-based models (ABMs) are widely used to predict how populations respond to changing environments. As the availability of food varies in space and time, individuals should have their own energy budgets, but there is no consensus as to how these should be modelled. Here, we use knowledge of physiological ecology to identify major issues confronting the modeller and to make recommendations about how energy budgets for use in ABMs should be constructed. 2. Our proposal is that modelled animals forage as necessary to supply their energy needs for maintenance, growth and reproduction. If there is sufficient energy intake, an animal allocates the energy obtained in the order: maintenance, growth, reproduction, energy storage, until its energy stores reach an optimal level. If there is a shortfall, the priorities for maintenance and growth/reproduction remain the same until reserves fall to a critical threshold below which all are allocated to maintenance. Rates of ingestion and allocation depend on body mass and temperature. We make suggestions for how each of these processes should be modelled mathematically. 3. Mortality rates vary with body mass and temperature according to known relationships, and these can be used to obtain estimates of background mortality rate. 4. If parameter values cannot be obtained directly, then values may provisionally be obtained by parameter borrowing, pattern-oriented modelling, artificial evolution or from allometric equations. 5. The development of ABMs incorporating individual energy budgets is essential for realistic modelling of populations affected by food availability. Such ABMs are already being used to guide conservation planning of nature reserves and shell fisheries, to assess environmental impacts of building proposals including wind farms and highways and to assess the effects on nontarget organisms of chemicals for the control of agricultural pests. Keywords: bioenergetics; energy budget; individual-based models; population dynamics.

Incorporating toxicokinetics into an individual-based model for more realistic pesticide exposure estimates: A case study of the wood mouse

The potential risk of agricultural pesticides to mammals typically depends on internal concentrations within individuals, and these are determined by the amount ingested and by absorption, distribution, metabolism, and excretion (ADME). Pesticide residues ingested depend, amongst other things, on individual spatial choices which determine how much and when feeding sites and areas of pesticide application overlap, and can be calculated using individual-based models (IBMs). Internal concentrations can be calculated using toxicokinetic (TK) models, which are quantitative representations of ADME processes. Here we provide a population model for the wood mouse (Apodemus sylvaticus) in which TK submodels were incorporated into an IBM representation of individuals making choices about where to feed. This allows us to estimate the contribution of individual spatial choice and TK processes to risk. We compared the risk predicted by four IBMs: (i) “AllExposed-NonTK”: assuming no spatial choice so all mice have 100% exposure, no TK, (ii) “AllExposed-TK”: identical to (i) except that the TK processes are included where individuals vary because they have different temporal patterns of ingestion in the IBM, (iii) “Spatial-NonTK”: individual spatial choice, no TK, and (iv) “Spatial-TK”: individual spatial choice and with TK. The TK parameters for hypothetical pesticides used in this study were selected such that a conventional risk assessment would fail. Exposures were standardised using risk quotients (RQ; exposure divided by LD50 or LC50). We found that for the exposed sub-population including either spatial choice or TK reduced the RQ by 37–85%, and for the total population the reduction was 37–94%. However spatial choice and TK together had little further effect in reducing RQ. The reasons for this are that when the proportion of time spent in treated crop (PT) approaches 1, TK processes dominate and spatial choice has very little effect, and conversely if PT is small spatial choice dominates and TK makes little contribution to exposure reduction. The latter situation means that a short time spent in the pesticide-treated field mimics exposure from a small gavage dose, but TK only makes a substantial difference when the dose was consumed over a longer period. We concluded that a combined TK-IBM is most likely to bring added value to the risk assessment process when the temporal pattern of feeding, time spent in exposed area and TK parameters are at an intermediate level; for instance wood mice in foliar spray scenarios spending more time in crop fields because of better plant cover.

Risk assessment of UK skylark populations using life-history and individual-based landscape models

Abstract: Following a workshop exercise, two models, an individual-based landscape model (IBLM) and a non-spatial life-history model were used to assess the impact of a fictitious insecticide on populations of skylarks in the UK. The chosen population endpoints were abundance, population growth rate, and the chances of population persistence. Both models used the same life-history descriptors and toxicity profiles as the basis for their parameter inputs. The models differed in that exposure was a pre-determined parameter in the life-history model, but an emergent property of the IBLM, and the IBLM required a landscape structure as an input. The model outputs were qualitatively similar between the two models. Under conditions dominated by winter wheat, both models predicted a population decline that was worsened by the use of the insecticide. Under broader habitat conditions, population declines were only predicted for the scenarios where the insecticide was added. Inputs to the models are very different, with the IBLM requiring a large volume of data in order to achieve the flexibility of being able to integrate a range of environmental and behavioural factors. The life-history model has very few explicit data inputs, but some of these relied on extensive prior modelling needing additional data as described in Roelofs et al.(2005, this volume). Both models have strengths and weaknesses; hence the ideal approach is that of combining the use of both simple and comprehensive modeling tools.

Hierarchical conditional random fields for parts based models matching

A parts based model is a parametrization of an object class using a collection of landmarks following the object structure. The matching of parts based models is one of the problems where pairwise Conditional Random Fields have been successfully applied. The main reason of their effectiveness is tractable inference and learning due to the simplicity of involved graphs, usually trees. However, these models do not consider possible patterns of statistics among sets of landmarks, and thus they sufffer from using too myopic information. To overcome this limitation, we propoese a novel structure based on a hierarchical Conditional Random Fields, which we explain in the first part of this memory. We build a hierarchy of combinations of landmarks, where matching is performed taking into account the whole hierarchy. To preserve tractable inference we effectively sample the label set. We test our method on facial feature selection and human pose estimation on two challenging datasets: Buffy and MultiPIE. In the second part of this memory, we present a novel approach to multiple kernel combination that relies on stacked classification. This method can be used to evaluate the landmarks of the parts-based model approach. Our method is based on combining responses of a set of independent classifiers for each individual kernel. Unlike earlier approaches that linearly combine kernel responses, our approach uses them as inputs to another set of classifiers. We will show that we outperform state-of-the-art methods on most of the standard benchmark datasets.

quantiNemo: an individual-based program to simulate quantitative traits with explicit genetic architecture in a dynamic metapopulation.

quantiNemo is an individual-based, genetically explicit stochastic simulation program. It was developed to investigate the effects of selection, mutation, recombination and drift on quantitative traits with varying architectures in structured populations connected by migration and located in a heterogeneous habitat. quantiNemo is highly flexible at various levels: population, selection, trait(s) architecture, genetic map for QTL and/or markers, environment, demography, mating system, etc. quantiNemo is coded in C++ using an object-oriented approach and runs on any computer platform. Availability: Executables for several platforms, user's manual, and source code are freely available under the GNU General Public License at http://www2.unil.ch/popgen/softwares/quantinemo.

INDISIM-YEAST, an individual-based model to study yeast

INDISIM-YEAST, an individual-based simulator, models the evolution of a yeast population by settingup rules of behaviour for each individual cell according to their own biological rules and characteristics. Ittakes into account the uptake, metabolism, budding reproduction and viability of the yeast cells, over aperiod of time in the bulk of a liquid medium, occupying a three dimensional closed spatial grid with twokinds of particles (glucose and ethanol). Each microorganism is characterized by its biomass, genealogicalage, states in the budding cellular reproduction cycle and position in the space among others. Simulationsare carried out for population properties (global properties), as well as for those properties that pertain toindividual yeast cells (microscopic properties). The results of the simulations are in good qualitativeagreement with established experimental trends.

Classifying general nonlinear force laws in cell-based models via the continuum limit

though discrete cell-based frameworks are now commonly used to simulate a whole range of biological phenomena, it is typically not obvious how the numerous different types of model are related to one another, nor which one is most appropriate in a given context. Here we demonstrate how individual cell movement on the discrete scale modeled using nonlinear force laws can be described by nonlinear diffusion coefficients on the continuum scale. A general relationship between nonlinear force laws and their respective diffusion coefficients is derived in one spatial dimension and, subsequently, a range of particular examples is considered. For each case excellent agreement is observed between numerical solutions of the discrete and corresponding continuum models. Three case studies are considered in which we demonstrate how the derived nonlinear diffusion coefficients can be used to (a) relate different discrete models of cell behavior; (b) derive discrete, intercell force laws from previously posed diffusion coefficients, and (c) describe aggregative behavior in discrete simulations.

Population based models of cortical drug response: insights from anaesthesia

A great explanatory gap lies between the molecular pharmacology of psychoactive agents and the neurophysiological changes they induce, as recorded by neuroimaging modalities. Causally relating the cellular actions of psychoactive compounds to their influence on population activity is experimentally challenging. Recent developments in the dynamical modelling of neural tissue have attempted to span this explanatory gap between microscopic targets and their macroscopic neurophysiological effects via a range of biologically plausible dynamical models of cortical tissue. Such theoretical models allow exploration of neural dynamics, in particular their modification by drug action. The ability to theoretically bridge scales is due to a biologically plausible averaging of cortical tissue properties. In the resulting macroscopic neural field, individual neurons need not be explicitly represented (as in neural networks). The following paper aims to provide a non-technical introduction to the mean field population modelling of drug action and its recent successes in modelling anaesthesia.

Using an individual-based model to select among alternative foraging strategies of woodpigeons: data support a memory-based model with a flocking mechanism

Population modelling is increasingly recognised as a useful tool for pesticide risk assessment. For vertebrates that may ingest pesticides with their food, such as woodpigeon (Columba palumbus), population models that simulate foraging behaviour explicitly can help predicting both exposure and population-level impact. Optimal foraging theory is often assumed to explain the individual-level decisions driving distributions of individuals in the field, but it may not adequately predict spatial and temporal characteristics of woodpigeon foraging because of the woodpigeons’ excellent memory, ability to fly long distances, and distinctive flocking behaviour. Here we present an individual-based model (IBM) of the woodpigeon. We used the model to predict distributions of foraging woodpigeons that use one of six alternative foraging strategies: optimal foraging, memory-based foraging and random foraging, each with or without flocking mechanisms. We used pattern-oriented modelling to determine which of the foraging strategies is best able to reproduce observed data patterns. Data used for model evaluation were gathered during a long-term woodpigeon study conducted between 1961 and 2004 and a radiotracking study conducted in 2003 and 2004, both in the UK, and are summarised here as three complex patterns: the distributions of foraging birds between vegetation types during the year, the number of fields visited daily by individuals, and the proportion of fields revisited by them on subsequent days. The model with a memory-based foraging strategy and a flocking mechanism was the only one to reproduce these three data patterns, and the optimal foraging model produced poor matches to all of them. The random foraging strategy reproduced two of the three patterns but was not able to guarantee population persistence. We conclude that with the memory-based foraging strategy including a flocking mechanism our model is realistic enough to estimate the potential exposure of woodpigeons to pesticides. We discuss how exposure can be linked to our model, and how the model could be used for risk assessment of pesticides, for example predicting exposure and effects in heterogeneous landscapes planted seasonally with a variety of crops, while accounting for differences in land use between landscapes.

Exploring community assembly through an individual-based model for trophic interactions

Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)

Trait contributions to fish community assembly emerge from trophic interactions in an individual-based model

Community ecology seeks to understand and predict the characteristics of communities that can develop under different environmental conditions, but most theory has been built on analytical models that are limited in the diversity of species traits that can be considered simultaneously. We address that limitation with an individual-based model to simulate assembly of fish communities characterized by life history and trophic interactions with multiple physiological tradeoffs as constraints on species performance. Simulation experiments were carried out to evaluate the distribution of 6 life history and 4 feeding traits along gradients of resource productivity and prey accessibility. These experiments revealed that traits differ greatly in importance for species sorting along the gradients. Body growth rate emerged as a key factor distinguishing community types and defining patterns of community stability and coexistence, followed by egg size and maximum body size. Dominance by fast-growing, relatively large, and fecund species occurred more frequently in cases where functional responses were saturated (i.e. high productivity and/or prey accessibility). Such dominance was associated with large biomass fluctuations and priority effects, which prevented richness from increasing with productivity and may have limited selection on secondary traits, such as spawning strategies and relative size at maturation. Our results illustrate that the distribution of species traits and the consequences for community dynamics are intimately linked and strictly dependent on how the benefits and costs of these traits are balanced across different conditions. © 2012 Elsevier B.V.