bifrost : an R package for scalable inference of phylogenetic shifts in multivariate evolutionary dynamics

High-dimensional comparative datasets, including geometric morphometric landmarks, functional traits, and other large trait datasets, are increasingly common in biology. When these datasets include a large number of traits relative to the number of taxa, they pose significant challenges for phylogenetic comparative analysis. In addition, evolutionary dynamics are often heterogeneous across phylogenies, challenging researchers to develop tools that can localize and account for such variation when investigating hypotheses of evolutionary change.

We present bifrost , an R package for detecting and characterizing shifts in multivariate trait evolution across phylogenetic trees. bifrost implements a stepwise greedy search over alternative macroevolutionary regime configurations on a phylogeny. Candidate shifts are proposed and assessed at internal nodes, accelerated with parallel model fitting where possible, and aggregated sequentially when they exceed a user-defined information-criterion acceptance threshold.

The underlying model is a scalar-rate multivariate Brownian motion process fit by generalized least squares using mvMORPH::mvgls [1]. Our framework also provides support estimates for individual shifts using information-criterion weights.

We illustrate the workflow using a fossil-tip-dated phylogeny and high-dimensional landmark data for early bony fish jaws (32,508 scalar coordinate values), and discuss tuning, outputs, and limitations. bifrost extends existing phylogenetic comparative frameworks for evolutionary analysis and provides a scalable pipeline for exploring the phylogenetic natural history of large multivariate datasets.