Inference in a metapopulation model via a composite-likelihood approximation. - INRAE - Institut national de recherche pour l’agriculture, l’alimentation et l’environnement Accéder directement au contenu
Communication Dans Un Congrès Année : 2018

Inference in a metapopulation model via a composite-likelihood approximation.

Résumé

Processes related to the spatio-temporal spread of pathogens in metapopulations are most often partially observed, and available data are usually incomplete, spread over time and heterogeneous. Moreover, the representation of this type of biological systems often leads to complex models. In this case, classical inference methods (i.e. maximum likelihood) are not usable because the likelihood function can not be specified. Bovine paratuberculosis (agent Mycobacterium avium subsp. paratuberculosis - Map) is a worldwide enzootic disease of economic importance whose screening in the field is diffcult due to its long incubation period and the low sensitivity of routine diagnostic tests. Our objective was to estimate key parameters of a multiscale dynamic model of Map spread from a longitudinal and spatial dataset collected in Brittany (Western France), using a specific approach taking into account accurately the characteristics of the census data, and to provide additional knowledge on the propagation of Map. Our approach is based on a stochastic mechanistic model of Map spread between dairy herds through animal trade movements. Comprehensive data on cattle movements in 12,857 dairy herds in Brittany and partial data on animal infection status (2,013 herds sampled from 2005 to 2013) were available. Inference was performed with a Monte-Carlo approximation of a composite likelihood coupled to a numerical optimization algorithm (Nelder-Mead Simplex-like). The six estimated key parameters of this model are: (i) the proportion of initially infected herds, (ii and iii) their infection level (distribution of within-herd prevalence), (iv) the probability of purchasing infected cattle from outside the metapopulation, (v) the indirect local transmission rate, and (vi) the sensitivity of the diagnostic test. Empirical identifiability was verified on simulated data. The optimization algorithm converged after appropriate tuning. Point estimates and profile likelihoods indicate a very large proportion (> 0.80) of infected herds with a low within-herd prevalence on average at the initial time (2005), a low risk of introducing an infected animal from outside ( 0.10) and a low sensitivity of the diagnostic test ( 0.25). Estimations of previously unknown key parameters provide new insights on Map spread at the regional scale, mainly showing a high prevalence in the number of infected herds, in agreement with qualitative opinions of experts. These estimates of previously unknown parameters provide new insights on Map status in Western France. The inference framework could easily be applied to datasets from other regions concerned by paratuberculosis and adapt to estimate key features of other spatio-temporal infection dynamics, most often imperfectly observed, especially for long-lasting endemic diseases. It is of particular interest when ABC-like inference methods fail due to difficulties in defining relevant summary statistics.
Fichier non déposé

Dates et versions

hal-02733573 , version 1 (02-06-2020)

Identifiants

  • HAL Id : hal-02733573 , version 1
  • PRODINRA : 443111

Citer

Gael Beaunée, Pauline Ezanno, Alain Joly, Pierre Nicolas, Elisabeta Vergu. Inference in a metapopulation model via a composite-likelihood approximation.. 11. European Conference on Mathematical and Theoretical Biology (ECMTB), Jul 2018, Lisbonne, Portugal. 882 p. ⟨hal-02733573⟩
17 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More