Ramsés H. Mena

Hierarchical Mixture Modeling With Normalized Inverse-Gaussian Priors

In recent years the Dirichlet process prior has experienced a great success in the context of Bayesian mixture modeling. The idea of overcoming discreteness of its realizations by exploiting it in hierarchical models, combined with the development of suitable sampling techniques, represent one of the reasons of its popularity. In this article we propose the normalized inverse-Gaussian (N–IG) process as an alternative to the Dirichlet process to be used in Bayesian hierarchical models. The N–IG prior is constructed via its finite-dimensional distributions. This prior, although sharing the discreteness property of the Dirichlet prior, is characterized by a more elaborate and sensible clustering which makes use of all the information contained in the data. Whereas in the Dirichlet case the mass assigned to each observation depends solely on the number of times that it occurred, for the N–IG prior the weight of a single observation depends heavily on the whole number of ties in the sample. Moreover, expressions corresponding to relevant statistical quantities, such as a priori moments and the predictive distributions, are as tractable as those arising from the Dirichlet process. This implies that well-established sampling schemes can be easily extended to cover hierarchical models based on the N–IG process. The mixture of N–IG process and the mixture of Dirichlet process are compared using two examples involving mixtures of normals.

Bayesian Nonparametric Estimation of the Probability of Discovering New Species

We consider the problem of evaluating the probability of discovering a certain number of new species in a new sample of population units, conditional on the number of species recorded in a basic sample. We use a Bayesian nonparametric approach. The different species proportions are assumed to be random and the observations from the population exchangeable. We provide a Bayesian estimator, under quadratic loss, for the probability of discovering new species which can be compared with well-known frequentist estimators. The results we obtain are illustrated through a numerical example and an application to a genomic dataset concerning the discovery of new genes by sequencing additional single-read sequences of cdna fragments. Copyright 2007, Oxford University Press.

Are Gibbs-Type Priors the Most Natural Generalization of the Dirichlet Process?

Discrete random probability measures and the exchangeable random partitions they induce are key tools for addressing a variety of estimation and prediction problems in Bayesian inference. Here we focus on the family of Gibbs–type priors, a recent elegant generalization of the Dirichlet and the Pitman–Yor process priors. These random probability measures share properties that are appealing both from a theoretical and an applied point of view: (i) they admit an intuitive predictive characterization justifying their use in terms of a precise assumption on the learning mechanism; (ii) they stand out in terms of mathematical tractability; (iii) they include several interesting special cases besides the Dirichlet and the Pitman–Yor processes. The goal of our paper is to provide a systematic and unified treatment of Gibbs–type priors and highlight their implications for Bayesian nonparametric inference. We deal with their distributional properties, the resulting estimators, frequentist asymptotic validation and the construction of time–dependent versions. Applications, mainly concerning mixture models and species sampling, serve to convey the main ideas. The intuition inherent to this class of priors and the neat results they lead to make one wonder whether it actually represents the most natural generalization of the Dirichlet process.Discrete random probability measures and the exchangeable random partitions they induce are key tools for addressing a variety of estimation and prediction problems in Bayesian inference. Here we focus on the family of Gibbs-type priors, a recent elegant generalization of the Dirichlet and the Pitman-Yor process priors. These random probability measures share properties that are appealing both from a theoretical and an applied point of view: (i) they admit an intuitive predictive characterization justifying their use in terms of a precise assumption on the learning mechanism; (ii) they stand out in terms of mathematical tractability; (iii) they include several interesting special cases besides the Dirichlet and the Pitman-Yor processes. The goal of our paper is to provide a systematic and unified treatment of Gibbs-type priors and highlight their implications for Bayesian nonparametric inference. We deal with their distributional properties, the resulting estimators, frequentist asymptotic validation and the construction of time-dependent versions. Applications, mainly concerning mixture models and species sampling, serve to convey the main ideas. The intuition inherent to this class of priors and the neat results they lead to make one wonder whether it actually represents the most natural generalization of the Dirichlet process.

A Bayesian nonparametric method for prediction in EST analysis

BackgroundExpressed sequence tags (ESTs) analyses are a fundamental tool for gene identification in organisms. Given a preliminary EST sample from a certain library, several statistical prediction problems arise. In particular, it is of interest to estimate how many new genes can be detected in a future EST sample of given size and also to determine the gene discovery rate: these estimates represent the basis for deciding whether to proceed sequencing the library and, in case of a positive decision, a guideline for selecting the size of the new sample. Such information is also useful for establishing sequencing efficiency in experimental design and for measuring the degree of redundancy of an EST library.ResultsIn this work we propose a Bayesian nonparametric approach for tackling statistical problems related to EST surveys. In particular, we provide estimates for: a) the coverage, defined as the proportion of unique genes in the library represented in the given sample of reads; b) the number of new unique genes to be observed in a future sample; c) the discovery rate of new genes as a function of the future sample size. The Bayesian nonparametric model we adopt conveys, in a statistically rigorous way, the available information into prediction. Our proposal has appealing properties over frequentist nonparametric methods, which become unstable when prediction is required for large future samples. EST libraries, previously studied with frequentist methods, are analyzed in detail.ConclusionThe Bayesian nonparametric approach we undertake yields valuable tools for gene capture and prediction in EST libraries. The estimators we obtain do not feature the kind of drawbacks associated with frequentist estimators and are reliable for any size of the additional sample.

A New Bayesian Nonparametric Mixture Model

We propose a new mixture model for Bayesian nonparametric inference. Rather than considering extensions from current approaches, such as the mixture of Dirichlet process model, we end up shrinking it, by making the weights less complex. We demonstrate the model and discuss its performance.

A Bayesian Nonparametric Approach for Comparing Clustering Structures in EST Libraries

Inference for Expressed Sequence Tags (ESTs) data is considered. We focus on evaluating the redundancy of a cDNA library and, more importantly, on comparing different libraries on the basis of their clustering structure. The numerical results we achieve allow us to assess the effect of an error correction procedure for EST data and to study the compatibility of single EST libraries with respect to merged ones. The proposed method is based on a Bayesian nonparametric approach that allows to understand the clustering mechanism that generates the observed data. As specific nonparametric model we use the two parameter Poisson-Dirichlet (PD) process. The PD process represents a tractable nonparametric prior which is a natural candidate for modeling data arising from discrete distributions. It allows prediction and testing in order to analyze the clustering structure featured by the data. We show how a full Bayesian analysis can be performed and describe the corresponding computational algorithm.

A flexible class of parametric transition regression models based on copulas: application to poliomyelitis incidence

This paper presents an extension of a general parametric class of transitional models of order p. In these models, the conditional distribution of the current observation, given the present and past history, is a mixture of conditional distributions, each of them corresponding to the current observation, given each one of the p-lagged observations. Such conditional distributions are constructed using bivariate copula models which allow for a rich range of dependence suitable to model non-Gaussian time series. Fixed and time varying covariates can be included in the models. These models have the advantage of straightforward construction and estimation for the analysis of time series and more general longitudinal data. A poliomyelitis incidence data set is used to illustrate the proposed methods, contrary to other researches’ conclusions whose methods are mainly based on linear models, we find significant evidence of a decreasing trend in polio infection after accounting for seasonality.

Dynamic density estimation with diffusive Dirichlet mixtures

We introduce a new class of nonparametric prior distributions on the space of continuously varying densities, induced by Dirichlet process mixtures which diffuse in time. These select time-indexed random functions without jumps, whose sections are continuous or discrete distributions depending on the choice of kernel. The construction exploits the widely used stick-breaking representation of the Dirichlet process and induces the time dependence by replacing the stick-breaking components with one-dimensional Wright-Fisher diffusions. These features combine appealing properties of the model, inherited from the Wright-Fisher diffusions and the Dirichlet mixture structure, with great flexibility and tractability for posterior computation. The construction can be easily extended to multi-parameter GEM marginal states, which include, for example, the Pitman--Yor process. A full inferential strategy is detailed and illustrated on simulated and real data.

On the construction of stationary AR(1) models via random distributions

We explore a method for constructing first-order stationary autoregressive-type models with given marginal distributions. We impose the underlying dependence structure in the model using Bayesian non-parametric predictive distributions. This approach allows for nonlinear dependency and at the same time works for any choice of marginal distribution. In particular, we look at the case of discrete-valued models; that is the marginal distributions are supported on the non-negative integers.

A time dependent Bayesian nonparametric model for air quality analysis

Air quality monitoring is based on pollutants concentration levels, typically recorded in metropolitan areas. These exhibit spatial and temporal dependence as well as seasonality trends, and their analysis demands flexible and robust statistical models. Here we propose to model the measurements of particulate matter, composed by atmospheric carcinogenic agents, by means of a Bayesian nonparametric dynamic model which accommodates the dependence structures present in the data and allows for fast and efficient posterior computation. Lead by the need to infer the probability of threshold crossing at arbitrary time points, crucial in contingency decision making, we apply the model to the time-varying density estimation for a PM 2.5 dataset collected in Santiago, Chile, and analyze various other quantities of interest derived from the estimate.