Research‎ > ‎

EcoSim: An ecosystem simulation

EcoSim is now open source! See the download page.

"A New Approach to Evolutionary Biology". International Innovation journal has published an article about our research project.

 The tree of life generated by a run of the simulation

Follow the weekly report of our virtual evolving ecosystem EcoSim's very long run. Week 19 - Time step 23000

Ecological modeling is a still growing field, at the crossroad between theoretical ecology, mathematics and computer science. Ecosystem models aim to characterize the major dynamics of ecosystems, in order to synthesize the understanding of such systems, and to allow predictions of their behavior. Because natural ecosystems are very complex (in terms of number of species and of ecological interactions) ecosystem models typically simplify the systems they are representing to a limited number of components. This simplification allows for the development of computer-aided ecosystem simulations that are tractable. One of the main interests of such ecosystem simulations is that they offer a global view of the evolution of the system, which is difficult to observe in nature. However, the scope of ecosystem simulations has always been limited by the computational possibilities of their time. Today, it is possible to run simulations that are more complex than what has ever been done before.

In the area of ecosystem simulation, individual-based modeling provides a bottom-up approach allowing for the consideration of the traits and behavior of individual organisms. Instead of modeling an ecosystem as a whole, individual-based models treat individuals as “unique and discrete entities”. By modeling organisms with varying characteristics (such as age, mating preferences and role in the ecosystem), the properties of the system that the individuals represent can begin to emerge from their interactions. While the use of individual behavior has been included in many models during recent decades, the individual-based modeling approach is receiving more consideration as the cost to purchase and operate a machine capable of running time-consuming simulations reduces. Yet, few attempts have been made to simulate a complete and complex ecosystem. An example of such a system is the platform Echo, which includes an evolutionary mechanism. However, the organisms in Echo are very simple, and are not provided with any behavior model. Another system studying long term evolution is the Avida. It nevertheless has some limitations such as: the individuals do not move and are quite limited in number, there is no explicit representation of species and more importantly there is a fixed fitness function which means that the system is in fact mostly an optimization process and therefore cannot be considered as a complex system. Other models, such as PolyWorld, Bubbleworld.Evo or Framsticks, have been proposed including more complex agents and behavioral models.  They use Artificial Neural Networks or system of learned rules to evolve the agent’s behavioral model during their life and by an evolutionary process. These systems have led to several interesting findings on complexity and structure of the network or the rules evolved by the system. However, these approaches are highly computationally expensive and only allow the implementation of a small population (few hundred) of agents. they are therefore more dedicated to investigate evolution of learning capacities than large scale evolutionary mechanisms involving population and species dynamics.

A run of the simulation EcoSim from time step 1 to 1660. Predators are in white and a random color is associated to every prey species.

The global objective of our work is to develop such a powerful predator-prey ecosystem simulation, called EcoSim, and to try to gain some knowledge about natural ecosystems thanks to it. We have created a generic platform able to simulate complex ecosystems with intelligent predator or prey agents interacting and evolving in a large and dynamic environment. We have chosen to implement an individual-based model, built upon the predator–prey paradigm. The novelty of our model stems from the fact that each agent behavior is modeled by a fuzzy cognitive map (FCM, a computational tool, similar to a neural network, based on a graph that represents interaction between concepts, such as emotions and desires, perception or action), and evolves during the simulation. The FCM of each agent is unique, and is the outcome of the evolution process going on throughout the simulation. The notion of species is also implemented, in a way that species emerge from the evolution of agents. To our knowledge, our system has been the first one allowing modeling the links between speciation and individual behavior.

Our simulation EcoSim produces a large amount of data that can be grouped in three categories: (i) some global statistics, such as the numbers of individuals and of species, or the quantity of resources available; (ii) some information relative to each species, such as its average FCM, the number of individuals it contains or their average fitness; (iii) some data relative to each agent, such as its level of energy, or its age. Thanks to this data, we plan to investigate macroevolutionary processes, such as the emergence of species, and specific ecological issues, such as species abundance patterns or invasive species. Our preliminary results have shown coherent behaviors of the whole simulation, with the emergence of strong correlation patterns observed also in field ecosystems

More videos of the simulation are available.

A detail description of our simulation is given in:

Can physical obstacles accelerate speciation?

The origin of species remains one of the most controversial and least understood topics in evolution. While it is widely accepted that complete cessation of gene-flow between populations due to long-lasting geographic barriers results in a steady, irreversible increase of divergence and eventually speciation, the extent to which various degrees of habitat heterogeneity influences speciation rates is less well understood. Here, we investigate how small, randomly distributed physical obstacles influence the distribution of populations and species, the level of population connectivity (e.g., gene flow) as well as the mode and tempo of speciation in EcoSim. We adapted EcoSim to allow fine tuning of the gene flow's level between populations by adding various numbers of obstacles in the world.

We compared three different configurations with different densities of obstacles. It is clear that the speciation rate and species diversity is directly proportional with the roughness of the physical environment. Our study also reveals that species are more spatially compact in the configurations with obstacles than in the reference world. Moreover, the continuous reduction in spatial distribution as the number of obstacles increases results in low levels of gene flow between sister species. Therefore, the rapid genomic divergence between species should be directly linked to the reduction of movement due to obstacles that result in low gene flow and rapid divergence between subdivided populations. We investigated several factors that could be involved in the increase of speciation rate, such as the individual’s behaviors, the spatial distribution of the species or the overall speed of evolution (increase in genetic divergence between sister species). We show that the faster divergence between populations and accelerated speciation cannot be explained by an increase of spatial separation during the initial stage of speciation or different behaviors of the individuals. We suggest that this is likely due to the significantly lower population sizes in obstacle configurations. This reduced size, results in more pronounced genetic drift and rapid differentiation between population that experience relatively low levels of gene flow.

It is well accepted that the effect of micro-geographic barriers (e.g., the raggedness of the environment) to maintain population cohesion and the genetic homogeneity of a species depends heavily on the intrinsic properties of the species (e.g., dispersal ability, intra- and inter- specific interactions). We suggest that the complex context of speciation can be better understood outside the framework of a classical geographical definition of speciation (e.g. sympatric, allopatric) by focusing on the complex interactions at the community level.

Speciation and extinctions are very important processes that influence the species composition of an ecosystem at a particular time and the long-term dynamics of ecological communities. Our approach allows testing the unified neutral theory of biodiversity and biogeography which suggests that the persistence of ecologically equivalent species in sympatry across relevant time scales might not depend strictly of complex niche differences. Our results suggest that factors affecting demographic stochasticity (e.g., factors shaping the density of individuals in a local area, the extent of the distribution of species in space) can influence speciation and extinction rates and ultimately the distribution of relative species abundance. Our approach has demonstrated its utility to model several important biological problems and it seems possible to modify it to represent many new ones.

This study has been published in:

Comparison of phylogenetic tree construction algorithms

Phylogenetic trees are constructed frequently in biological research to provide an understanding of the evolutionary history of the organisms being studied. Often, the actual phylogenetic tree is unknown and the phylogenetic tree constructed is an estimate. There are many methods of phylogenetic tree construction which fall into two main categories: distance-based methods and character-based methods. To test the accuracy of these methods, it is necessary that the system being studied is one for which the actual phylogenetic tree is known. EcoSim is an ecosystem simulation in which predator and prey agents possessing a complex behavioral model can interact, evolve and speciate. In this experiment, we used EcoSim to test the accuracy of the three main distance-based phylogenetic tree construction methods, when constructing a single tree and when performing phylogenetic bootstrapping. Since EcoSim provides data regarding speciation events, we were able to construct the actual phylogenetic trees from this data. We then performed the UPGMA, Neighbor-Joining, and Fitch-Margoliash methods at various time-steps and used symmetric distance as a metric to compare the topologies of the actual and estimated trees. On average, trees contained nearly 30 taxa. We found that the Fitch-Margoliash method with bootstrapping performed slightly better than the other methods, however no method constructed trees in which more than 50% of the partitions were correct.

This study has been published in:

  • Scott R., Gras R., Comparing Distance-Based Phylogenetic Tree Construction Methods Using An Individual-Based Ecosystem Simulation, EcoSim, The Thirteenth International Conference on the Synthesis and Simulation of Living Systems - Artificial Life 13, 105-110, 2012.

Global behavior: a chaotic process

The purpose of this project is the examination of the stochastic and deterministic behavior of signals that are produced by an ecosystem simulation. We are interested to understand how complex and predictable our EcoSim simulation is. The variations of the time series associated to ecosystem simulation are the result of complex simulation mechanisms. We have for example, studied the properties of the time series representing the variation of the number of preys, the number of prey species, the number of predators and the number of predator species.

To understand how close our system is to the random or chaotic processes, we examine whether a chaotic behavior exists in these signals. To enforce our result, we use four different methods: Higuchi fractal dimension, correlation dimension, largest Lyapunov exponent, P&H method. For each of them, in order to obtain a statistically significant evaluation, we apply the surrogate test method on 24 samplings of the considered data. According to the results obtained after applying these different methods, all of them providing clear predictions, we can conclude that behavior of simulation is chaotic. 

It is important to show the existence of a complex behavior in our simulation because any attempt to model a realistic system need to have the capacity to generate patterns as complex as the ones that are observed in real systems. In this regards, we will be interested to analyze other versions of the current simulation or even other different simulations. This will give us the possibility to better understand what the important factors are that leads to complex systems with behavior showing important level of regularity but still very hard to predict. Using the same process it is also possible to examine other time series generated by our simulation. It can be useful to determine among the huge amount of data generated by the simulation the ones that are the more interesting to analyze more in deep.

We also examined the multifractal behavior of signals that are produced by EcoSim.
We focused on the time series corresponding to the variation of the number of prey and predator individuals and the individuals' positions. We also applied our analysis to a random walk version of the simulation and compare it with the behavioral model simulation. In the random walk version all the action such as eat, hunting, reproduction are replaced by a random movement action. The distribution of movements and the size of the world are the same as in the original simulation. In each time step, a specific number of predator and prey die and birth. In order to still have a realistic system, the Lotka-Volterra competition model with density dependence has been used to compute these numbers.
With this comparison it has been shown that multifractal characteristic happened only in behavioral model simulation. Individual spatial distribution could be fitted reasonably well with multifractal models. It showed that the data generated by our simulation present the same kind of multifractal properties as the ones observed in real ecosystems. It is also another important confirmation of the capacity of our simulation to model complex and realistic large scale systems. This fact also suggests the applicability of Rényi dimensions for the characterization of individuals' population distribution and for the simulation and modeling of empirical data by means of algorithms, such as iterated function systems, capable of generating such kind of measures

This study has been published in:
  • Golestani A., Gras R., Identifying Origin of Self-Similarity in EcoSim, an Individual-Based Ecosystem Simulation, Using Wavelet-based Multifractal Analysis,  Proceedings of the World Congress on Engineering and Computer Science 2012 (WCECS 2012), San Francisco, 1275-1282.
  • Golestani A., Gras R., Multifractal Phenomena in EcoSim, a Large Scale Individual-Based Ecosystem Simulation, ICAI (International Conference on Artificial Intelligence) 2011, Las Vegas, USA, 991-997.
  • Golestani A., Gras R., Regularity Analysis of an individual-based Ecosystem Simulation, Chaos: An Interdisciplinary Journal of Nonlinear Science, 2010.
  • Majdabadi Farahani Y., Golestani A., Gras R., Complexity and Chaos Analysis of a Predator-Prey Ecosystem Simulation, The second international conference on Advanced Cognitive Technologies and Applications, 2010, Lisbon, Portugal, November.

Species abundance patterns

To further the validation of our simulation, we have compared the ecological patterns emerging from our ecosystem EcoSIm with those observed in natural ecosystems. We have focused the study on species abundance patterns (number of prey or predator species containing a given number of individuals) because they are a key component of macroecological theories. To analyze these patterns, we have chosen to use Fisher’s logseries, since it is one of the most classical models of species abundance distribution. These comparisons have been performed by testing the goodness-of-fit between an observed distribution and the one calculated by using the logseries. We have defined a very general concept of good fit that can be statistically evaluated on any kind of distribution. 

The following results, that are well established in the ecological literature, are also observed in the communities generated by our simulation: (i) the logseries presents a good fit to the species abundance distributions of relatively small samples; (ii) it fails to do so for large samples and complete community enumerations; (iii) the logseries performs better on species-rich than on species-poor communities. Even though the logseries does not provide a good fit for large samples, the species abundance distribution patterns observed in our communities are similar to those observed in large samples of natural communities. Thus, at any level in sample size, our simulation gives coherent results in terms of relative species abundance, when compared with classical ecological results.

One result is particularly interesting, for it allows making predictions. If the total number of species in the community is known, one can identify what type of curve will be produced. Conversely, and more interestingly, if one considers a community for which only a few samples are available, and if these samples allow to identify what kind of curve could be constructed (by using interpolation techniques), one can obtain a lower bound or an upper bound of the total number of species in the community.

To sum up, we can say that the confrontation of the data generated by our simulation with field data has strongly validated our simulation in terms of relative species abundance. On the other hand, the study of our communities has extended ecological field results. For example, we have discovered a dichotomy between species-rich and species-poor communities (with a threshold of 240 species) in terms of the evolution of the goodness-of-fit to the logseries according to sample size.

This study has been published in:

Speciation mechanism

While the presence of individual-based models continues to rise, to our knowledge, there has been very little detailed study on the simulation of various speciation methods within an evolving individual-based ecosystem.
As a mechanism for clustering individuals into similar groups, we implemented in EcoSim a 2-means clustering technique designed to allow for (1) the splitting of an existing species S into S1 and S2, and (2) the clustering of individuals that initially belonged to S into one of either S1 or S2 . We have then analyzed how our speciation mechanism produce data compatible with real observations.

Computing the Spearman's rank correlation coefficient between the time series representing the variation of number of individuals and number of species we observed
, as it is expected, a strong correlation. In fact, the strong positive correlation between the number of prey and the number of prey species is at a maximum at a distance of approximately 50 time steps. This suggests that, as the quantity of prey individuals increases, so does the quantity of prey species 50 time steps later.

Introduced in 1943 by Sewall Wright, “Isolation by distance” is a biological theory
that suggests a positive correlation between physical distances and genetic differences. Subsequent authors, including Kimura and Weiss (in 1964), Nagylaki (in 1976), and Slatkin (in 2007) have continued to study this phenomenon, the last of which demonstrated that on samples of genes from two populations, it is possible to identify isolation by distance.
For every pair of individuals in a species, (I1, I2), measuring the physical distance and genetic distance between I1 and I2 demonstrates some evidence of isolation by distance. Visualizing the physical location of individuals within the world helps us to identify a relationship between the physical location of individuals and their genetic similarity.
The Figure depicts the physical locations of individuals in prey species 286 and 425 – before and after splitting. It can be seen that the new cluster of genetically similar individuals, which form the new prey species, are also physically located near each other.

We shown also the effect of genetic drift by measuring the genetic distance between the average genome of two species just after a speciation event and 1000 time steps letter. After 1000 time steps there is no more overlap between the species and their average genome is much more dissimilar than just after speciation.

This study has been published in:
  • Aspinal A., Gras R., K-Means Clustering as a Speciation Method within an Individual-Based Evolving Predator-Prey Ecosystem Simulation, Lecture Notes in Computer Science, 6335, 318-329, 2010, Toronto, August, IEEE International Conferences on Active Media Technology.

Prediction of species' extinction

In the course of gradual long-drawn-out evolution, there has been an innumerable number of species; not all of them could manage to live until now and some of them lived longer than the others. For a species to survive, its individuals have to reproduce and tolerate environmental conditions. Population extinctions which are a milestone of ecology have applications in conservation biology, biological control, epidemiology, and genetics. Although much research has been carried out in population extinctions and populations’ persistence, this phenomenon is still an open question. There are many factors in population extinctions that can be classified into the three main realms of Demography, Genetics and Environment. Demographic factors that include population variability, initial population size and migration are impacted by population growth, reproduction rate, and individual’s lifespan. Genetic factors correspond to a shortage of genetic variations, which can be caused by a decrease in fitness due to inbreeding depression. Finally, factors such as habitat quality, habitat fragmentation, and environmental stressors have a major role in population extinctions. The effects of most of these factors depend on interactions with other factors and conditions, which impede the careful scrutiny of each factor separately. In real life, it is difficult to identify or compute an exact effect of these factors separately and it is even harder to do it for altogether. Simulation techniques can be a good alternative to inspect factors together. Our predator-prey ecosystem simulation EcoSim enables us to investigate different aspects of life in a multi-level food chain simulation. The predators act as a pressure factor on preys and can be seen as an environmental stress. The preys eat grass, which availability is based on a spatial diffusion model leading to a dynamically changing environment for the prey.

  Accuracy TN rate
TP rate
 ROC Area
 Train 92% 95% 84% 0.96
 Test 94.9% 95% 84% 0.96
 Validation 90.3% 91% 76% 0.92

We work on the notion of prediction of species’ extinctions based on several species’ features. For this study, we use different attributes of species, linked to demographic and genetic factors, called features. These features were gathered from individuals that belong to distinctive species existing in the world. We investigate several machine learning methods to build a predictor of close future species extinctions. This approach allows us to work on numerous factors simultaneously. Based on this model we set up three experiences with different species’ features on two datasets. Results confirmed the impact of demographic and genetic factors on prediction of species extinction and showed that very good predictor can be built. We demonstrated that a combination of these factors can improve the prediction’s accuracy. In addition, we validated our results by applying our predictor learned on a run of the simulation to prediction species's extinction on other runs. The predictions obtained were a little less accurate than the ones obtained using the same run for testing but were still above 90% accuracy. It means that the predictors discovered are quite general and that it exists some specific caracteristics of species than can be used to predict their extinction even in different conditions.

In a next step, we want to focus on  correlation and dependency between features. For this purpose, we have to work on the analysis of features’ interactions and on the extraction of biologically significant rules. These rules will help to reveal the priority and relation between features and provide some insight about the biological mechanisms involved in species’ extinction.

This study has been published in:
  • Hosseini M, Gras R., Md Sina, 'Prediction of Imminent Species’ Extinction in EcoSim, International Conference on Agents and Artificial Intelligence 2012, Portugal, 318-323.

Population spatial distribution 

Spatial distribution of population carries important information both in real life and simulated environment
for analyzing various aspects of species or any group of individuals. One of the applications is prediction of extinction of a population. When individuals of a population are dying, there spatial distribution either globally or locally, starts to decrease; since it has a relationship with number of living individuals of the population. Also it was shown that genetically similar individuals of a population tend to live closer to each other. Computation of spatial distribution requires exact location information of individuals of interest, which is almost always impossible in real life. But in a computer simulation such as Individual-Based ecosystem simulation, this becomes possible. Simulators usually simulates a continuous world where there is no starting or end point, which for the sake of implementation simplicity is made discontinuous. In this work we develop a simple linear time algorithm that uses circular statistics to compute spatial distributions in our Individual-Based ecosystem simulation EcoSim. The algorithm allows to compute spatial center of a population in a torus like world and since we have the center of the distribution we can compute other statistical measures.  There are a number of possible extensions of this study; for example, it will be interesting to study how the spatial center of a population moves over time with respect to other populations. Another possibility is to find different clusters of individuals of same species.

This study has been published in:
  • Sina M., Gras R., Computation of population spatial distribution in individual-based ecosystem simulation, IEEE ALIFE 2011, Paris, France, April, in press.

Speciation Prediction based on Spatial Distribution and Spatiotemporal Information

Speciation is the division of one single species into two or more genetically distinct ones. This process extends through time and leads to a hierarchal tree of historical relationship between species. Two steps are entailed in speciation. First a new population should be established which could be in the same habitat or completely separated of the main population depending on which speciation mechanism is involved. Second a reproductive isolation should occur, due to a kind of barrier like different habitats, physical barrier, etc., to reduce or prevent gene flow between organisms of the different species. Therefore, the geographical and spatial distribution of individuals in one species is a leading phenomenon for speciation. It has been shown that an increase in physical distance between individuals imply that the genetic distance between them also increases. If the genetic distance between individuals of the same population is too high reproductive isolation will occur and leading to a speciation event. Consequently, increasing the physical distance between individuals increases the probability of speciation. However, considering spatial distribution metrics alone is not enough for studying speciation. Because it is a continuous ongoing process, the current spatial distribution of a species is not necessarily a reliable index of the species' historical distribution during its life time.  Losos et al. mentioned three evidences showing that the present spatial distribution of a species is greatly different from the one at its creation time. Therefore, observing species during its whole life time is also important to understand and eventually predict speciation.

Although there many factors are involved in speciation, in this study we want to answer to the questions such as how spatial and spatiotemporal patterns influence speciation? Are spatial distribution metrics effective in speciation prediction? Which metrics are important and in what extent? Is spatiotemporal information of one species helpful in speciation prediction? Again which of them are more important?
For answering such questions, we have applied machine learning techniques on the data generated by Ecosim to evaluate if spatial distribution and spatiotemporal information of species can predict splitting of species. If we could predict speciation by using this information, it means that they have impact on species splitting. We are also interested to extract predictive rules on speciation based on spatial and spatiotemporal information that could help to understand this complex phenomenon.

We used 14 measures to extract the spatial (S) and spatio-temporal (ST) information and applyied oversampling technique to build classifiers. We obtained very good results for the test set coming from the same run as the learning set. It also comes out good results for the test sets coming from different runs showing that classifier can extract general rules about speciation. For all datasets S, ST, S+ST, we also observed that the classifier performance increases when the number of species contained in the learning set increases. It means that giving more example of speciation events, even if they come from the same run, make the predictor more generic, which in turn means that some generic traits exist in our simulation that characterize the speciation events.   This is highly important for the potentiality of our approach to discover some information useful for real prediction. Finally, we noticed that spatial information of individuals in Ecosim has tremendous effect on speciation prediction, as it has also been observed in real ecosystems, while spatiotemporal information can improve it in some extent. For future work, we will study more in detail the results of speciation prediction and extract some important rules involved in speciation. It is also possible to work on other information of species like their genome or mating factors, to give better prediction for speciation.

This study has been published in:
  • Mashayekhi M., Gras R., Investigating the Effect of Spatial Distribution and Spatiotemporal Information on Speciation using Individual-Based Ecosystem Simulation, Journal of Computing, 2(1), 98 – 103.
  • Mashayekhi M., Gras R., Speciation Prediction based on Spatial Distribution and Spatiotemporal Information from an  Individual-Based Ecosystem Simulation, Advanced Topics in Artificial Intelligence 2011, 56-62.

Correlation between Genetic Diversity and Fitness

Genetic diversity serves as a way for populations to adapt to changing environments. With more variation, it is more likely that some individuals in a population will possess variations of alleles that are suited for the environment. Those individuals are more likely to survive to produce offspring bearing that allele. The population will continue for more generations because of the success of these individuals. In summary, genetic diversity strengthens a population by increasing the likelihood that at least some of the individuals will be able to survive major disturbances, and by making the group less susceptible to inherited disorders. Many biological studies showed that decreased population genetic diversity can be associated with declines in population fitness. However, population learn also from its environment by selecting the individuals with highest fitness. This driving force is opposite to the previous one and leads to unstable equilibrium value for genetic diversity. Because overall population diversity affects both short-term individual fitness and long-term population adaptive capacity, there is a need to develop an empirical quantitative understanding of the relationship between population genetic diversity and population viability.

EcoSim gives us the chance to study the relation between species genetic diversity and species fitness, not only in certain environmental conditions and at specific time like done in biological studies, but also through evolution. In EcoSim the environment changes from one place to another and from a time step to another. Individuals that evolve in different parts of the world have different information stored in their genome about the environment they evolve in. Furthermore, as we model a predator-prey system, we have co-evolution. This means that the strategies (behaviors) of each kind are continuously changing trying to adapt to the other kind. Thus there are many factors affecting the genetic diversity and fitness and controlling values of correlation between them. At every time step we calculate entropy and fitness for all existing species. In order to investigate their possible correlations, we first begin by calculating the Spearman’s rank cross correlation for all prey species between their genetic diversity and their fitness. The Spearman measure ranks two sets of variables and tests for a linear relationship between the variables’ ranks. A perfect Spearman correlation of +1 or -1 occurs when each of the variables is a perfect monotone function of the other.

We found very high correlation both negative and positive between entropy and fitness. In order to validate our correlation results and further understand the reasons behind these results we built a classifier to predict the correlation class variable, positive or negative correlation, based on set of properties describing the species (for example the number of individuals in the species, their average energy level, the average number of reproduction events, the spatial dispersal...). The interest of building this classifier is first to see if some specific species properties can predict the current evolutionary behavior of a species (that is if it is learning from the environment or increasing its diversity to be able to react to a future change in the environment). It can also help to understand what are the factors and conditions that affect the evolutionary behavior. We found high accuracy for classification which proves the interest of our genetic diversity measure and its correlation with fitness. In addition, we used feature extraction to find the best features affecting the correlation values. We showed how these extracted features are similar to the factors affecting genetic diversity and fitness in community ecology. The similarity between results of five different runs of the simulation proves the stability of the simulation and the generality of our findings. This study allows us to shows that the relation between genetic diversity and fitness changes based on time and other features such as reproduction rate, population size and spatial dispersal. In the future we will investigate more about the values of the features and which values lead to negative or positive correlation which would have a great impact on community ecology domain. More in depth study about the evolution of different specific genes and its affect in the general behavior of individuals is also our research interest. Epistasis, measuring how one gene affect another or how several genes ”work” together, is also an important aspect worth studying.

This study has been published in:

  • Khater M., Salehi E., Gras R., Correlation between Genetic Diversity and Fitness in a Predator-Prey Ecosystem Simulation, 24th Australian Joint Conference on Artificial Intelligence, Perth, Australia, 2011, LNAI 7106, 422-431.
  • Khater M., Gras R., Adaptation and Genomic Evolution in EcoSim, 12th International Conference on Adaptive Behaviour, LNAI 7426, 219-229, 2012.

Projects in early stage

Modeling infectious diseases has been a major concern for the last century. We develop EcoDemics; which extends our ecosystem simulation to model the spread of infectious diseases. The dynamics along with the spatial distribution of agents impact disease dynamics. We study the effect of spreading a directly transmitted infectious disease among the prey agents. We implement a disease model and analyze the effect of different control strategies and study their impact in reducing the effect and duration of an epidemic. Among various control strategies, vaccination has proved a powerful defense against infectious diseases. We focus on vaccination as a mitigation technique and investigate the effect of varying different parameters in vaccinating procedure.

We are also testing the two main biological models for conditions leading to sympatric speciation: the direct model and the indirect model for assortative mating, i.e. for the preference to mate with like individuals. (i) In the indirect model, a change in the ecology of species (e.g. food specialization) affects the mating time or location, and assortative mating may evolve, as mating takes place largely among individuals sharing ecological traits. In our simulation, by adding another food resource for the prey, and creating species having different initial food preferences, we can determine whether mating preferences appear, by looking for relations between the probability of mating and individual distances. (ii) In the direct model for assortative mating, selection is supposed to operate directly on mating preferences, when the genes responsible for a trait influencing mating preferences also affect survival or fecundity. We can test that as well, given that our simulation allows to compute individual fitness.

We are also studying the effect of a non-uniform world. We have modified the model to allow a non-uniform distribution of the food resource for the preys. We are interested to understand what is the impact of such distribution on the speciation events and what kind of equilibrium in individual spatial distribution can be reached. In our first investigation, we use a circular distribution of food. Then we investigated other more complex distributions, like star or square distributions.

A run from time step 1 to 1500 of the simulation EcoSim with a star distribution of grass. Grass is in green, predators are in white and a random color is associated to every prey species.