Section: New Results

A method to combine combinatorial optimization and statistics to mine high-throughput genotyping data

Participants : Julie Hamon, Clarisse Dhaenens, Julien Jacques (MODAL)

In the context of genomic analysis (collaboration with Genes Diffusion), dealing with high-throughput genotyping data, the objective of our study is to select a subset of SNPs (single nucleotide polymorphisms) explaining a trait of interest. We propose a method combining combinatorial optimization and statistics to extract a subset of interesting SNPs. The combinatorial part aims at exploring in an efficient way the large search space induced by the large number of possible subsets and statistics are used to evaluate the selection. We propose a first method based on an ILS (iterated local search) and using a regression. Three criteria used to evaluate the quality of the regression are compared. One of them (the k-fold validation) shows better performance. We also compare this approach to classical statistical approaches on simulated datasets. Results are promising as the proposed approach outperforms most of these statistical approaches [51] .