Zero-inflated Poisson regression models for QTL mapping applied to tick-resistance in a Gyr x Holstein F2 population

Abstract

Now a days, an important and interesting alternative in the control of tick-infestation in cattle is to select resistant animals, and identify the respective quantitative trait loci (QTLs) and DNA markers, for posterior use in breeding programs. The number of ticks/animal is characterized as a discrete-counting trait, which could potentially follow Poisson distribution. However, in the case of an excess of zeros, due to the occurrence of several noninfected animals, zero-inflated Poisson and generalized zero-inflated distribution (GZIP) may provide a better description of the data. Thus, the objective here was to compare through simulation, Poisson and ZIP models (simple and generalized) with classical approaches, for QTL mapping with counting phenotypes under different scenarios, and to apply these approaches to a QTL study of tick resistance in an F2 cattle (Gyr × Holstein) population. It was concluded that, when working with zero-inflated data, it is recommendable to use the generalized and simple ZIP model for analysis. On the other hand, when working with data with zeros, but not zero-inflated, the Poisson model or a data-transformation-approach, such as square-root or Box-Cox transformation, are applicable.

Modeling zero‐modified count and semicontinuous data in health services research Part 1: background and overview

“…Silva et al . proposed a ZIP model for quantitative trait loci mapping. More recently, Kleinke and Reinecke developed multiple imputation methods for “missing at random” zero‐inflated count data.…”

“…For times-series analysis, Hasan and Sneddon [44] developed first-order autoregressive and moving average ZIP models. Silva et al [45] proposed a ZIP model for quantitative trait loci mapping. More recently, Kleinke and Reinecke [46] developed multiple imputation methods for "missing at random" zero-inflated count data.…”

Bayesian interval mapping of count trait loci based on zero‐inflated generalized Poisson regression model

“…However, all the above work did not involve QTL mapping. For QTL mapping of zero-inflated count traits, Cui and Yang (2009) used the EM (Expectation-Maximization) algorithm and Silva et al (2011) applied the ZIP regression model. More recently, Moghimbeigi (2015) showed that the two-part zero-inflated negative binomial (ZINB) regression model can be used to map QTLs with count trait.…”