Address: LIRMM CNRS UMR 5506, 860 rue de St Priest 34095 Montpellier cedex 5
I design probabilistic models of evolution and algorithms to infer their parameters from the analysis of molecular, fossil and/or spatial data. I created and am still developping the software package PhyML (for Phylogenetics through Maximum Likelihood) which serves as a basis to implement my research outputs.
Trained as a biologist/statistician, I am working as a CNRS research scientist in the computer science department of the LIRMM in Montpellier, France. I was also lucky to work for the Department of Statistics at the University of Auckland between 2007 and 2015.
Accounting for spatial sampling patterns in Bayesian phylogeography S. Guindon, N. De Maio.
Proceeding of the National Academy of Sciences, USA. 118(52), 2022. We propose a new statistical approach to accommodate for preferential
sampling in phylogeography. This new technique distinguishes between the case where the
density of samples in particular areas reflects the underlying population density and the
case where it does not. Read this "research highlight" in Nature Computational Science.
Sampling bias and model choice in continuous
phylogeography: getting lost on a random walk A. Kalkauskas, U. Perron, Y. Sun, N. Goldman, G. Baele, S. Guindon, N. De Maio.
PLOS Computational Biology. 17 (1), e1008561. 2021. In this article, we study the impact that spatial sampling schemes have on the inference of model
parameters under standard approaches in continuous phylogeography.
Rates and rocks: strengths and weaknesses of
molecular dating methods. S. Guindon.
Frontiers in Genetics. 11. 2020.
I give here an overview of some of the technical aspects of molecular dating methods, including
models of rate variation along lineages and the different ways to calibrate a dating
analysis using fossil data.
Accounting for ambiguity in ancestral sequence reconstruction. A. Oliva,
S. Pulicani, V. Lefort, L. Brehelin, O. Gascuel & S. Guindon. Bioinformatics. 35: 4290-4297. 2019.
We propose a new criterion that better accomodates for ambiguity in ancestral (nucleotide or
protein) sequence reconstruction. In particular, more than one nucleotide or amino-acid are
inferred whenever multiple states have similar marginal posterior probabilities. The
proposed approach is computationally efficient and does not rely on arbitrarily set tuning
Accounting for calibration uncertainty: Bayesian
molecular dating as a ``doubly intractable'' problem. S. Guindon. Systematic
Biology. 67: 651-661. 2018. I describe a technique for molecular dating that accomodates for
uncertainty in the placement of calibration constraints as defined by an expert. The method
relies on the so-called ``exchange algorithm'' for sampling from doubly intractable
distributions Cover of July issue!
Demographic inference under the coalescent in a
spatial continuum. S. Guindon, H. Guo, D. Welch. Journal of Theoretical Population
Biology. 111: 43–50. 2016. We describe a method that fits a structured coalescent model
assuming that individuals are scattered on a continuum rather that distributed in discrete
demes. We show that the density of the population and the rate of dispersal of
individuals can be inferred simultaneousy from the analysis of geo-referenced genetic
sequence using this technique.
Modeling competition and dispersal in a statistical phylogeographic
framework. L. Ranjard, D. Welch, M. Paturel, S. Guindon. Systematic
Biology. 63:743-752. 2014. We describe a model where the probability for a species to colonize an empty island
at some point during the course of evolution is the same at
that of an occupied one only if species do not compete with each other. Using simulations, we show that these probabilities can indeed be
estimated from geo-referenced genetic sequences.
Performance of standard and stochastic branch-site models for detecting positive selection amongst coding sequences
A. Lu, S. Guindon. Molecular Biology and Evolution. 31: 484-495. 2014. We
describe the performance of the stochastic branch-site model (see S. Guindon et al. PNAS. 101:12957-12962. 2004.) in terms of type-I and power for detecting
positive selection in coding sequences. Results indicate that this approach
is more suited than the standard branch-site model (sensu codeml) in cases
where it is not
known a priori which lineages may have evolved under positive selection.
From trajectories to averages: an improved description of the heterogeneity of
substitution rates along lineages. S. Guindon. Systematic
Biology. 62:22-34. 2013. Assuming that the evolution of the substitution rate at each position along a sequence is
a realization of a (geometric) Brownian process, the rate averaged
over a given time interval is approximately gamma distributed. This study shows that ignoring the stochasticity of
average substitution rates leads to poor estimates of important evolutionary parameters. The proposed
approach also provides an efficient implementation of the
covarion model that does not require augmentation of the state space.