Laurent Serlet

Statistical algorithms for the analysis of deleterious genetic mutations

We present algorithms for model selection and parameter estimation concerning deleterious genetic mutations. Three models are considered: single gene mutation, double cross-effect mutations or no genetic cause. Each of these models include unknown parameters that must be estimated simultaneously. Available data are phenotypes along family pedigrees but no genotypic data. We compare classical fit methods based on statistical summaries of the data and a neural network approach. We show the performance of our algorithms on simulated datasets of reasonable size. We also consider real data concerning breast/ovarian cancer.

Oncogenetic pedigrees: Relation between design and ability to predict mutation

We consider the risk of a disease caused by the presence within the genome of one deleterious mutation or two interacting mutations. For an individual, the probability of being mutated/doubly mutated can be estimated by knowing the phenotype of the other members in the family pedigree. We study the performance of this process as a function of the size, the shape of the family tree and the parameters of the model. We carry out simulations using the parameters pertaining to breast/ovarian cancer in BRCA-mutated families.