Modèles de mutation : étude probabiliste et estimation paramétrique

Abstract : Mutation models are probabilistic descriptions of the growth of a population of cells, where mutationsoccur randomly during the process. Data are samples of integers, interpreted as final numbers ofmutant cells. These numbers may be coupled with final numbers of cells (mutant and non mutant) or a mean final number of cells.The frequent appearance in the data of very large mutant counts, usually called “jackpots”, evidencesheavy-tailed probability distributions.Any mutation model can be interpreted as the result of three ingredients. The first ingredient is about the number of mutations occuring with small probabilityamong a large number of cell divisions. Due to the law of small numbers, the number of mutations approximately follows aPoisson distribution. The second ingredient models the developing duration of the clone stemming from each mutation. Due to exponentialgrowth, most mutations occur close to the end of the experiment. Thus the developing time of arandom clone has exponential distribution. The last ingredients represents the number of mutant cells that any clone developing for a given time will produce. Thedistribution of this number depends mainly on the distribution of division times of mutants.One of the most used mutation model is the Luria-Delbrück model.In these model, division times of mutant cells were supposed to be exponentially distributed.Thus a clone develops according to a Yule process and its size at any given time follows a geometric distribution.This approach leads to a family of probability distributions which depend on the expected number of mutations and the relative fitness, which is the ratio between the growth rate of normal cells to that of mutants.The statistic purpose of these models is the estimation of these parameters. The probability for amutant cell to appear upon any given cell division is estimated dividing the mean number of mutations by the mean final number of cells.Given samples of final mutant counts, it is possible to build estimators maximizing the likelihood, or usingprobability generating function.Computing robust estimates is of crucial importance in medical applications, like cancer tumor relapse or multidrug resistance of Mycobacterium Tuberculosis for instance.The problem with classical mutation models, is that they are based on quite unrealistic assumptions: constant final number of cells,no cell deaths, exponential distribution of lifetimes, or time homogeneity. Using a model for estimation, when thedata have been generated by another one, necessarily induces a bias on estimates.Several sources of bias has been partially dealed until now: non-exponential lifetimes, cell deaths, fluctuations of the final count of cells,dependence of the lifetimes, plating efficiency. The time homogeneity remains untreated.This thesis contains probabilistic and statistic study of mutation models taking into account the following bias sources:non-exponential and non-identical lifetimes, cell deaths, fluctuations of the final count of cells, plating efficiency.Simulation studies has been performed in order to propose robust estimation methods, whatever the modeling assumptions.The methods have also been applied to real data sets, to compare the results with the estimates obtained under classical models.An R package based on the different results obtained in this work has been implemented (joint work with Rémy Drouilhetand Stéphane Despréaux) and is available on the CRAN. It includes functions dedicated to the mutation models and parameter estimation.The applications have been developed for the Labex TOUCAN (Toulouse Cancer).
Document type :
Theses
Complete list of metadatas

Cited literature [102 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01631149
Contributor : Abes Star <>
Submitted on : Wednesday, January 17, 2018 - 8:46:07 AM
Last modification on : Saturday, December 29, 2018 - 1:14:01 AM
Long-term archiving on : Monday, May 7, 2018 - 3:57:52 PM

File

MAZOYER_2017_archivage.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01631149, version 2

Collections

Citation

Adrien Mazoyer. Modèles de mutation : étude probabiliste et estimation paramétrique. Bio-informatique [q-bio.QM]. Université Grenoble Alpes, 2017. Français. ⟨NNT : 2017GREAM032⟩. ⟨tel-01631149v2⟩

Share

Metrics

Record views

256

Files downloads

222