Mixture models of truncated data for estimating the number of species - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2010

Mixture models of truncated data for estimating the number of species

Résumé

Metagenomics goes beyond DNA sequencing by tackling communities of microorganisms in their natural environment. Previously, each microbial strain needed to be cultured before sequencing. Applying DNA sequencing directly to the sample has revealed the great diversity of the microbial populations in soil, sea water or the intestinal flora. Even though many new species can be studied, many more remain unobserved in the collected data. Estimating the total number of microbial species in the biological sample and their abundance distribution is key to determining the number of sequencing runs needed. In the standard model introduced by Fisher et al. (1943), each species contribute a Poisson-distributed number of individuals to the dataset, with a species-specific abundance parameter. Unobserved species are those that contribute zero individuals. Mixture models provide flexible models for the distribution of the abundance parameters. Following Bunge and Barger (2008), we use a truncated mixture model of geometric distributions. We propose to perform parameter estimation in the Bayesian framework with a variational algorithm, Beal and Ghahramani (2003). In this work, the number of components is not selected and we use Bayesian model averaging to combine the estimates from all considered models. In particular, the variational framework provides an efficient way of computing the weights for each model.
Fichier non déposé

Dates et versions

hal-01197572 , version 1 (11-09-2015)

Identifiants

  • HAL Id : hal-01197572 , version 1
  • PRODINRA : 51618

Citer

Sébastien Li-Thiao-Te, Jean-Jacques Daudin, Stephane Robin. Mixture models of truncated data for estimating the number of species. 19. International Conference on Computational Statistics, Aug 2010, Paris, France. ⟨hal-01197572⟩
43 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More