Abstract

One major challenge of using the phylogenetic comparative method (PCM) is the analysis of the evolution of interrelated continuous and discrete traits in a single multivariate statistical framework. In addition, more intricate parameters such as branch-specific directional selection have rarely been integrated into such multivariate PCM frameworks. Here, originally motivated to analyze the complex evolutionary trajectories of group size (continuous variable) and social systems (discrete variable) in African subterranean rodents, we develop a flexible approach using approximate Bayesian computation (ABC). Specifically, our multivariate ABC-PCM method allows the user to flexibly model an underlying latent evolutionary function between continuous and discrete traits. The ABC-PCM also simultaneously incorporates complex evolutionary parameters such as branch-specific selection. This study highlights the flexibility of ABC-PCMs in analyzing the evolution of phenotypic traits interrelated in a complex manner.

Highlights

  • Phylogenetic comparative methods (PCMs) provide a powerful statistical framework for investigating the patterns and processes of trait evolution (Felsenstein 1985; Harvey & Pagel1991; Nunn 2011; Garamszegi 2014)

  • The idea of the threshold model was originally developed in quantitative genetics by Wright (1934) to understand how multiple underlying genetic loci contribute to categorical traits such as the number of digits in guinea pigs

  • The threshold model assumes an unobservable continuous trait called ‘liability’ that underlies a discrete trait of interest

Read more

Summary

Introduction

Phylogenetic comparative methods (PCMs) provide a powerful statistical framework for investigating the patterns and processes of trait evolution (Felsenstein 1985; Harvey & Pagel1991; Nunn 2011; Garamszegi 2014). The recent development of PCMs permits analyses of biologically interrelated discrete and continuous variables in a single multivariate statistical framework (Table 1). The development of such multivariate PCMs is crucial for two reasons. The state of the discrete trait of interest is determined by whether the liability trait value is below or above a particular threshold. It is convenient to assume that the discrete trait is determined by a simple threshold (which can be treated as a probit link function in a framework of a phylogenetic generalized linear mixed model, PGLMM; Hadfield 2015; see Ives and Garland 2014), it is unclear if the assumption is always biologically valid. It is desirable for researchers to be able to assume other forms of latent functions (Fig. 1)

Methods
Results
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call