A Ratio-Based Method for Estimating an Unknown Number of Classes

• Activity Number: 334
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Amy Willis*+ and John Bunge
• Companies: Cornell University and Cornell University
• Keywords: species richness ; microbial ecology ; biodiversity ; nonlinear regression ; capture-recapture ; characterization of distributions

Abstract

We propose a method and implementation for estimating the total number of classes in a population. While the method is general, it is best suited to high-diversity microbial datasets, characterised by a large singleton count which is often attributed to errors in the DNA sequencing process. The classical diversity estimation approach models the sample taxa counts as independent Poisson random variables, the means of which are an i.i.d. sample from some mixing distribution. While this model is the foundation of every existing parametric method in this problem, it tends to fit poorly. We present the first explicit departure from the mixed Poisson model, a method for its implementation, and an R package. We draw on theory of characterization of distributions and find that ratio-based models fit microbial frequency datasets vastly better than existing models. The method gives sensible estimates of the number of unobserved species, sensible standard errors, and most importantly, believable and well-fitting models. Model selection techniques provide evidence that the mixed Poisson model is in fact not appropriate to the analysis of high-diversity datasets.