I am currently postdoc at ISEM, LIRMM and at the Computational Biology Institute in Montpellier, working with CÚline Scornavacca, Fabio Pardi and Vincent Berry. I am interested in methodological aspects of phylogenetic networks. Before, I was postdoc at the French National Institute for Agricultural Research (INRA) in Toulouse, working on genomic selection with Brigitte Mangin. I focused on Ridge regression, and its prediction when the number of SNPs (i.e. regressors) is larger than the number of individuals. Previously, I was postdoc in CÚcile AnÚ's lab, working on phylogenetics at the statistics department of the University of Wisconsin-Madison (USA). This area of research covers applied probability, statistics, computer science and biology. I completed my PhD at the Laboratory of Statistics and Probability (LSP) of Paul Sabatier's university, Toulouse. My supervisors were Jean-Marc Azais (LSP) and Jean-Michel Elsen (INRA). The research area of my PhD was on mathematical methods for Quantitative Trait Locus detection. I worked on selective genotyping experiments and also on the empirical process defined by the famous Interval mapping of Lander and Botstein.
Asymptotic Statistics / Mixture models / Statistical Genetics / Gaussian Processes / Chi-Square Processes
Computational Biology / Phylogenetics / Random Trees / Reconciliation between Gene Trees and Species Trees /
High-dimensional data analysis
I was a member of the stat/math phylogenetic group in Madison. This interdisciplinary group is composed of biologists, statisticians and mathematicians : David Baum, CÚcile AnÚ, Bret Larget, Sebastien Roch and students. The main goal is to propose new statistical methods to reconstruct evolutionary trees. Since more and more data from the genome are available, it becomes more and more interesting to study evolution events responsible for the diversity of genomes. Most of the methods involved in phylogenetics are Bayesian, since we have to deal with a huge parameter space. We analyze thousand of gene families and try to infer the phylogenetic tree for each of these families. Since this work requires a huge amount of computer resources, we work in collaboration with Bill Taylor in order to use the technology offered by HTCondor, the product of years of research by the Center for High Throughput Computing in the Department of Computer Sciences at UW Madison. HTCondor is used by Dreamworks in order to make movies and was also used in the recent discovery of the Higgs boson, a new particle in physics. We are particularly interested in evolutionary events called Whole Genome Duplications (WGD), i.e. an event which creates an organism with additional copies of the entire genome of a species. WGD can be found in bananas, tomatoes, flowering plants ... Download our softwares for reconstructing gene trees and detecting WGD !
First, QTL means Quantitative Trait Locus. It is any locus responsible for the variation of a quantitative trait. On this picture, you can see a few plamtrees which come from Lanzarote (Canary Islands). We may wonder why the first palmtree is largely taller than the others. There is maybe a QTL responsible for this particularity. In order to locate the QTL, people scan the genome of the palmtrees and do a statistical test at each position. That is the principle of interval mapping (Lander and Botstein 89). Since a lot of tests are performed on the genome, we are confronting to multiplicity problem. The work of Benjamini and Hochberg (95) on the false discovery rate is a way to deal with this problems of multi testing. Another approach is to study the stochastic process defined by all the correlated tests performed along the genome. I study the asymptotic properties of the corresponding empirical process. The Matlab pictures on the top of this web page gives a summary on the asymptotic properties of the empirical process. On the left, is represented the correlation whereas on the right, is given the mean function of the empirical process, that is to say the signal which contains information about the QTL. On the other hand, genotyping was very expensive in the past. That's why Lebotwitz (1987) proposed to genotype only individuals with extreme phenotypes : it is selective genotyping. I study the asymptotic properties of such a design and study also the empirical process under selective genotyping. Note that selective genotyping is still very useful nowdays, even if thousands of genetic markers are now available. We can increase the signal by genotyping only extreme individuals. Download our softwares to compute the quantiles of the maximum of Gaussian processes and Chi Square processes!