|
Organizers |
Modeling the invasion of Hawthorn at Porters Pass, NZ
by
Boris Baeumer
University of Otago
Coauthors: Agnes Radl
Hawthorn was introduced to Porters Pass in 1924. In 1983 and 2006 all Hawthorn trees in the area had their age estimated. We use this unique data set to test different (non-local) invasion of species models.
Date received: October 30, 2008
Empirical challenges in the evolution of the human genome
by
Gill Bejerano
Stanford University
I will outline briefly a modern view of the Human Genome highlighting the following points of potential interest: Extreme genomic sequence conservation---how surprising is it really? The discrepancy between observed short term and inferred long term dispensability of mammalian genomic DNA; Our growing appreciation for the potential complexity of the vertebrate gene regulatory code.
Date received: November 3, 2008
Poisson methods for modeling extinction and mutation
by
Peter D Drummond
Swinburne University of Technology
Coauthors: Alexei J. Drummond, T. G. Vaughan
We present an explicit unified stochastic model of fluctuations in population size due to random birth, death, density-dependent competition, mutation and environmental fluctuations. Stochastic dynamics provide insight into small populations, including processes such as extinction, that cannot be correctly treated by deterministic methods. We give exact analytical and simulation-based results for extinction times of our stochastic model without mutation with comparisons of the effects of environmental noise and intrinsic demographic stochasticity. Several methods - the discrete master equation approach, an exact mapping to a Fokker-Planck equation (the Poisson method), and stochastic equations are employed - showing they are precisely equivalent. We also calculate approximate extinction times using a steepest descent method, and demonstrate the ecological survival merit of using `unselfish' reduced birth rates, instead of `selfish' competition to control population size. This model can readily be extended to accommodate metapopulation structure and genetic variation, for example in viral populations. It thus represents a step towards a microscopic synthesis of population dynamics and population genetics.
Date received: November 3, 2008
Robust consensus methods for summarising phylogenetic trees
by
Barbara Holland
Allan Wilson Centre, Massey University
Coauthors: Barbara Keil
In phylogenetics consensus methods take as input a set of phylogenetic trees (on identical label sets) and attempt to identify where these trees agree. One popular method, majority-rule consensus displays all edges that appear in more than half the trees. Another popular approach, Adams consensus, preserves all the rooted triples that are displayed by all the input trees. For ``noisy'' data sets with large numbers of taxa both these methods can produce unresolved trees. This talk describes an attempt to modify the Adams consensus method by using ideas from the majority-rule approach to create a consensus method that is more robust to noisy data.
Date received: October 7, 2008
Reconstructing the evolutionary past of polyploid species: new combinatorial results
by
Katharina Huber
School of Computing Sciences, University of East Anglia, UK.
Coauthors: Martin Lott, Vincent Moulton, and Andreas Spillner.
Polyploid organisms are very common within plants but are also well documented within some animal groups. Essentially, such organisms arise when two organisms from different species hybridize giving rise to progeny with (possibly multiple) copies of their parents genome. The importance of polyploids for e.g. food production makes the development of methodology and algorithmic tools for reconstructing their evolutionary past an important albeit challenging task.
Recently a phylogenetic network reconstruction technique was introduced with this in mind. Its starting point is some kind of consensus over a set of multiply labeled trees (essentially rooted graph-theoretical trees in which every vertex of degree at most 2 is labeled but two distinct labeled vertices may have the same label) each supported by e.g. some gene. In this talk we will present recent results concerning the construction of such trees.
Date received: October 23, 2008
Fast phylogeny reconstruction through learning of ancestral sequences
by
Radu Mihaescu
UC Berkeley
Coauthors: Cameron Hill, Satish Rao, Alex Jaffe
Phylogenetic tree reconstruction is the task of recovering the topology of an evolutionary tree T from the evolved sequences at its leaves X. We present an algorithm which recovers the full phylogenetic tree T from logarithmic sized leaf sequences under the Cavender-Farris model of evolution with edge lengths under the Ising model phase transition: the probability of mutation along each edge is smaller than p0=(√(2)-1)/(2 √(2)).
Our work builds on recent progress by Daskalakis, Mossel, and Roch (DMR) who settled a conjecture of M. Steel by providing an O(n10) worst case running time algorithm achieving the same asymptotic results. The main advantages of our approach reside in the asymptotically optimal running time O(n2 log(n)) and the ability to provide partial topological information when some edges violate the length restriction.
We are able to circumvent the need for a-priori knowledge on lower and upper bounds on the edge lengths. Rather, we infer an edge length reliability interval from the size of the available sequences and proceed to recover the extremal components of the sub-forest of T given by edges falling in this reliability interval. In the case of trees with edges under the phase transition, the sequence length required for total reconstruction matches that of DMR.
Our methods are motivated by an intuitive minimum spanning tree framework. Similarly to DMR, we rely heavily on a method of Mossel for reconstructing sequences at the ancestral nodes of the tree with a bounded probability of error, in itself a very important problem in computational biology.
Date received: September 26, 2008
Testing hypotheses of treelikeness in genomic datasets
by
Vincent Moulton
School of Computing Sciences, University of East Anglia
Coauthors: Jo Dicks, Katharina Huber, George Savva
A common assumption of a phylogenetic analysis is that the evolutionary history of a dataset is best represented by a bifurcating tree. However, evolutionary processes including genetic recombination, horizontal gene transfer and allopolyploidy have been identified that give rise to datasets that cannot be represented adequately in this way. Our growing understanding of the importance of these events has led to the development of several mathematical and graphical representations of non-treelike evolution, phylogenetic networks. These approaches identify, within a dataset, conflicts with the hypothesis of treelike evolution. However, it is also known that apparent conflicts can arise from random variation in a dataset or through an incorrectly specified model of evolution rather than any underlying reticulate evolutionary process, so it is often difficult to intepret the results of a phylogenetic network analysis. In this talk, we address the problem of deciding whether or not a tree is adequate to describe the evolutionary signal a dataset contains. In particular, we introduce and discuss a simple test to assess the statistical significance of a non-treelike signal identified by a particular phylogenetic network tool, the NeighborNet.
Date received: October 15, 2008
Finding the trees in Darwin's forest
by
Lior Pachter
UC Berkeley
Coauthors: Robert Bradley, Nicolas Bray, Colin Dewey and Ariel Schwartz
The problem of determining homology among multiple related biological sequences, known as the alignment problem, is arguably the fundamental problem in comparative genomics. Accurate alignment is essential for both functional and evolutionary genomics studies. We explain how the problem of determining homology at the nucleotide level can be interpreted as finding the trees in "Darwin's forest", and focus on the tractability of the problem. In this letter, we argue that many recent negative results emphasizing uncertainty in alignment are misleading in that they confound uncertainty in the choice of model, uncertainty in alignment given a model, and error due to heuristics used for inference. We explain how hidden Markov models for pairwise alignment can be extended to provide effective models for multiple alignment, and show that these models indicate little uncertainty in alignment of both unrelated sequences and of orthologous sequences from related species. Moreover, we discuss an algorithm that provides an efficient approach to finding the alignments with highest expected accuracy. Together, these results provide a path to the removal of lingering doubts about the accuracy of multiple alignments.
Date received: September 2, 2008
Decision theory for saving species
by
Hugh Possingham
The University of Queensland, Maths and Ecology
Conservation science is booming, however much of its progress is hampered by a lack of quantitative thinking and tools. In this talk I will pose and solve three problems in conservation science that involve economic constraints. First we look at the problem of allocating resources between threatened species, some which may, or may not, be interdependent. Second we look at the problem of allocating resources to different actions across the globe. Third, with time, I will look at a problem of managing a threatened or harvested population where one of the objectives is learning.
Date received: October 21, 2008
Understanding frequency-dependent selection using the pairwise-interaction model
by
Hamish G Spencer
Allan Wilson Centre for Molecular Ecology & Evolution, University of Otago
Frequency-dependent selection (FDS) occurs when the fitnesses of the different genetic types of organisms in a population depend on their relative frequencies. In predator-prey situations, for instance, a rare prey type may be overlooked because it does not fit the predator’s search image. Hence, being rare confers selective advantage over commoner types and, clearly, by favouring rare genetic types, this sort of FDS has the potential to preserve genetic variability in a population. Observations of natural populations show that genetic variation is, in fact, ubiquitous. Why this is so is a central, unanswered problem in population-genetic theory, although FDS is often invoked as a plausible heuristic.
This talk examines the ability of a general model of FDS, the pairwise-interaction model, to maintain genetic variation. We also explore some mathematical properties of this model, showing, for example, why selection does not generally lead to populations with their mean fitness at a maximum.
Date received: August 24, 2008
A new method for tackling the stochastic dynamics of viral infection
by
Timothy G. Vaughan
Swinburne University of Technology
Coauthors: Peter D. Drummond and Alexei J. Drummond
The dynamics of viral infection are intrinsically stochastic in nature, and inter-dependent processes such as the infection of cells and the subsequent production of virions by those cells lead to the development of statistical correlations between the various sub-populations involved. Such correlations form an integral part of the population dynamics, but are often difficult to calculate for realistic numbers of cells and virions using standard Monte Carlo techniques. In this talk, we will demonstrate how stochastic models of viral infection can be tackled using the Poisson representation - an exact means of expressing a discrete birth/death master equation in terms of a diffusion process. By comparing the numerical results thus obtained with those calculated using Gillespie's exact algorithm, we will assess the validity and effectiveness of this approach, and discuss its potential application to the investigation of the mutation-driven evolutionary dynamics of single-host viral infections.
Date received: November 3, 2008