Abstracts for the "Evolutionary modelling in palaeontology" symposium are listed below.
Slicing the stratigraphic cake: the effects of time subsample variation in disparity-through-time analysis
Natalie Cooper1, Thomas Guillerme2,3
1Natural History Museum, London
2Imperial College London
3University of Queensland
Disparity-through-time analyses are commonly conducted in palaeontology. These analyses investigate how the morphological diversity of a group changes through time, and are used to answer key questions, such as how do mass extinction events influence diversity beyond loss of species richness, and whether some groups out-competed another by filling their ecological niche. These analyses are common but we don't always carefully consider the details of the methods used. Yet how you choose to measure disparity and time subsample your data can have consequences for your conclusions, sometimes changing them dramatically. Using a variety of palaeontological datasets, we show how decisions about time binning your data can change the results of your analyses. We present a new method for time slicing datasets taking into account the group of interest's phylogeny and several evolutionary models, implemented in dispRity, a new R package that allows you to use this method and more, in a flexible and reproducible manner. We also present results for groups with no available phylogeny.
Simulating evolution in space and time
Russell Garwood1, Mark D. Sutton2, Chris Knight1, Guillaume Gomez1, Alan R.T. Spencer2
1University of Manchester
The identification of evolutionary patterns in deep time is frustrated by biases in the fossil record. Geology, taphonomy, and collection all alter what ancient life can tell us about evolution. These filters can be both avoided and studied through in silico simulation, i.e. by using a simplified, abstracted model to explore the patterns and processes of evolution. We present such a model, coded into complementary software packages, EvoSim and EvoTree that incorporate many aspects of biological evolution. They allow experimental investigation on palaeontological timescales with sizeable populations of individuals, each with an explicit genome (millions of digital organisms over millions of generations). The software: operates with (or without) geography and environmental variation; allows asexual and sexual reproduction; identifies species (which arise naturally); and documents true evolutionary trees. We present data created by the simulation, and demonstrate patterns also observed in microbial evolutionary experiments. We can use these models to address key evolutionary questions: how evolutionary patterns/rates are impacted by rates of environmental change; the evolution of sexual reproduction; and the dynamics of mass extinctions. These simulations allow a range of hypotheses developed in palaeontology, and evolutionary biology, to be tested, and new ones to be developed.
Modelling biotic interactions using data from the fossil record
Lee Hsiang Liow1
1University of Oslo
Biotic interactions such as predation, disease, competition, symbiosis, mutualism, sexual selection are important processes that influence ecology and evolution in major ways. However, neither direct nor indirect evidence of such interactions are easy to observe or infer from the fossil record. I present two disparate approaches I use to approach the problem of describing and inferring fossilized biotic interactions. The first is a tool for causal inference (in the Granger sense) that involves linear stochastic differential equations that we have applied to diversification time series of bivalves and brachiopods over the Phanerozoic. I present evidence that bivalves deterred brachiopods evolutionarily by “allowing” brachiopods to diversify when bivalve extinctions are high. The second is an empirical system in which we could directly observe outcomes of intra- and interspecific competition. This system involves encrusting bryozoans which compete for space on hard substrates; we have collected data on thousands of interactions in order to understand the causes and consequences of competition through more than 2 million years of bryozoan evolution. I demonstrate how we modelled competition using ordinal regression, with a twist, to figure out which organismal traits were responsible for competitive outcomes.
Journeys through discrete character morphospace
1University of Leeds
The rich morphological data set that is the fossil record has been summarised in the form of discrete character-taxon matrices since at least the middle of the 20th century. Such matrices have been used to understand evolutionary tempo, infer phylogenetic trees, and measure morphological diversity (disparity). Historically these have been seen as separate endeavours, but they may be better understood within a common framework: a discrete-character phylomorphospace. Such a space is hyperdimensional, with points representing nodes of a phylogenetic tree (tips and reconstructed ancestors) and the branches linking them reflecting the journeys morphological evolution has taken. Such spaces thus offer the potential to better understand how patterns of morphological diversity emerge. Here I outline several difficulties in generating such spaces. These include: 1) the non-Euclidean nature of most morphological distances, 2) difficulties in incorporating various uncertainties (including phylogenetic), and 3) problems associated with visualising high dimensional spaces. In addition there are multiple routes to generating such phylomorphospaces, the main difference being whether ancestors are reconstructed pre- or post-ordination. I explore this major division using an empirical data set of theropod dinosaurs and show that some routes are likely to generate phylomorphospaces that better reflect phylogenetic than morphological distances.
Evolution and Earth Systems: modeling population-level processes on palaeontological scales
P. David Polly1
Evolution is fundamentally a population-level process in which variation, drift, and selection interact to produce patterns of change (or stasis) that are both spatial and temporal. Palaeontology provides empirical data on the outcomes of evolution, which sometimes point to unexpected macroevolutionary patterns that arise from interactions between evolutionary processes and large-scale phenomena like tectonics, climatic change, or clade coevolution. Because population and palaeontological data often differ in granularity, it can be difficult to relate processes at these two scales to one another. Computational modelling can be used as a tool for extrapolating population level processes over palaeontological scales, allowing in silico experimental tests of competing evolutionary explanations for the same paleontological outcomes. Here I review basic properties of population level processes – variation, drift, selection, and population size – and I discuss the relationship between their spatial and temporal dynamics. I then show how typical evolutionary models used in palaeontology and phylogenetics (Brownian motion, Ornstein-Uhlenbeck, and directional) incorporate the temporal component of these dynamics, but not the spatial. Finally, I show using computational modelling that large scale Earth system processes can create evolutionary outcomes that depart from basic population-level notions from these standard macroevolutionary models.
Using evolutionary models to assess the accuracy of phylogenies estimated with Bayesian, Maximum-Likelihood, and Parsimony methods
Mark Puttick1,2, Joseph O’Reilly2, Davide Pisani2, Phil Donoghue2
1University of Bath
2University of Bristol
Placing fossils in the tree of life is a fundamental aim of evolutionary biology, but it is unclear which phylogenetic methods are most accurate when estimating phylogenies based on morphological data. Molecular phylogeneticists have generally abandoned parsimony as it exhibits statistically undesirable properties and is less accurate than probabilistic phylogenetic models. In palaeontology, equal weights and implied weights parsimony are widely used, but recent research suggests probabilistic alternatives, mainly the Bayesian Mk model, may provide greater accuracy. Here we review these recent debates to show that parsimony generally performs poorly in comparison to probabilistic models such as the Bayesian Mk model. Many of these differences are due to Bayesian model incorporating uncertainty not accommodated in equal weights or implied weights parsimony. However, when this uncertainty is incorporated in parsimony approaches, probabilistic methods still achieve higher accuracy. Finally, we show that even when data are simulated to be biased towards parsimony, and implied weights parsimony in particular, Bayesian estimation still achieves higher accuracy in datasets rich in homoplasy. The simplistic Mk model of evolution used in Bayesian phylogenetics is more accurate than parsimony alternatives, and unlike parsimony, it offers future scope to model the complexities of morphological data.
How different is reality from mathematical perfection in taxonomy?
2Queen’s University Belfast
There are intrinsic mathematical patterns in nature. Species are natural units that are formed from branching phylogenetic processes that also have a mathematical structure. We should be able develop a set of general principles that describe global patterns of species groups, like genera, or families. Understanding such patterns would lend considerable power to “taxonomic surrogacy”, the common practice of substituting genus or family-level names where species identification is impractical. Assessing the error introduced by taxonomic surrogacy could also improve comparisons of diversity patterns in the fossil record to the living biota. But some higher taxa are “natural” or monophyletic groups, while others are mixed or paraphyletic melting pots awaiting taxonomic revision. Finally, the use of species group designations are potentially different in living and fossil taxa. All of these issues can be addressed through simulation approaches, in silico approximations of both large scale phylogenetic scenarios that underpin the evolution of species groups, combined with simulation of an “idealised” taxonomic practice. Clarity and confidence in fundamental patterns for taxonomy based on a robust null model – species-poor genera are very common, large genera are very rare – may provide tools that accelerate species discovery in fossil and living groups.