top of page

My current research interests include artificial intelligence, machine learning, data mining and combinatorial optimization.
 

Past Topics

 

Molecular phylogeny      

Correction of dissimilarity matrices, for phylogeny reconstruction, continuation of a collaboration with the LIMOS, at Clermont-Ferrand

 

Research of motifs common to several sequences

studied during my PhD thesis and further

 

Comparative genomics

Identification of putative strong transcription promoters, a collaboration with the Biotechnology Laboratory of the University of Nantes

 

Modeling of gene regulatory networks

Model transformation and integration of temporal constraints,

a collaboration between two Computer Science Research Laboratories of Nantes (IRRCyN and LINA)

 

Inference of haplotypes from unphased genotypes

Inference of the complete information borne by each of two paired chromosomes (i.e. the phases), for all individuals in a population, relying on (unphased) genotypes

 

Large-scale inference of missing data, for genetical data

Exploitation of local similarities within groups of individuals, within a given population

​

Simulation of realistic genetical data

In GWASs, the statistical power of any novel proposal to detect susceptibility variants has to be assessed. For this purpose, one has to simulate genotypes and phenotypes under the hypothesis of association, that is with a dependence between some genetic marker(s) and the studied phenotype.

​

Recovery of epistatic patterns of susceptibility in complex diseases

Epistasy defines the case when some genetic markers have no individual effect on a given phenotype, whereas their combination has one. 

​

Recent Topics

 

Modeling of dependences between event traces and multivariate time series from a cohort 
The aim is to detect regime changes and to model the evolution dynamics within each regime in a semi-supervised context (i.e., with both known and unknown states).

​

Fast recognition of contextualized multidimensional patterns in sets of annotated multivariate time series

This topic is investigated to allow future health professionals to practice on a digital patient, in the simulated operating room, thanks to reactive scenarios. Each time the learner triggers a medical action, the digital patient's physiological parameters are modified based on the records of (real) observed patients most similar to the digital patient, in their recent multivariate trajectories. This process is iterated from action to action, under strong real-time constraints.

 

 

Synthetic representation to describe clusters of event trajectories
For instance, the challenge is to obtain succinct representations for clusters of patients described by temporal health-care data.

 

​

Consensus recommendation for a group
The challenge consists in selecting items to collectively satisty at best the preferences  of a group of subjects.
​
Random forests with latent variables
This model was proposed to handle feature selection in highly correlated data.
​
Feature subset selection in high-dimensional settings
Focus on methods falling into the following categories: random forest, Bayesian inference, ant colony optimization, Markov blanket learning.
​
Modeling of high-dimensional and spatially correlated data

A novel class of probabilistic graphical models, the forest of latent tree models, was proposed, and a scalable learning algorithm was designed.


Advanced models in machine learning for advanced GWAS strategies

An association study aims at identifying associations between genetic markers (e.g. Single Nucleotide Polymorphisms - SNPs) and phenotype (e.g. a disease, plant resistance to a bioagressor). Genome wide association studies pose formidable challenges.​

bottom of page