Gametocytes infectiousness to mosquitoes: variable selection using random forests, and zero inflated models

Robin Genuer 1, 2 Isabelle Morlais 3 Wilson Toussile 1
2 SELECT - Model selection in statistical learning
Inria Saclay - Ile de France, LMO - Laboratoire de Mathématiques d'Orsay, CNRS - Centre National de la Recherche Scientifique : UMR
Abstract : Malaria control strategies aiming at reducing disease transmission intensity may impact both oocyst intensity and infection prevalence in the mosquito vector. Thus far, mathematical models failed to identify a clear relationship between Plasmodium falciparum gametocytes and their infectiousness to mosquitoes. Natural isolates of gametocytes are genetically diverse and biologically complex. Infectiousness to mosquitoes relies on multiple parameters such as density, sex-ratio, maturity, parasite genotypes and host immune factors. In this article, we investigated how density and genetic diversity of gametocytes impact on the success of transmission in the mosquito vector. We analyzed data for which the number of covariates plus attendant interactions is at least of order of the sample size, precluding usage of classical models such as general linear models. We then considered the variable importance from random forests to address the problem of selecting the most influent variables. The selected covariates were assessed in the zero inflated negative binomial model which accommodates both over-dispersion and the sources of non infected mosquitoes. We found that the most important covariates related to infection prevalence and parasite intensity are gametocyte density and multiplicity of infection.
Liste complète des métadonnées
Contributeur : Robin Genuer <>
Soumis le : lundi 21 février 2011 - 15:21:35
Dernière modification le : jeudi 9 février 2017 - 15:53:39
Document(s) archivé(s) le : dimanche 22 mai 2011 - 02:58:20


Fichiers produits par l'(les) auteur(s)




Robin Genuer, Isabelle Morlais, Wilson Toussile. Gametocytes infectiousness to mosquitoes: variable selection using random forests, and zero inflated models. [Research Report] RR-7497, INRIA. 2011. <inria-00550980v3>



Consultations de
la notice


Téléchargements du document