Efficient estimation of evolutionary rates by covariance aware regression

Abstract

Shared ancestry among individuals results in correlated traits and these dependencies need to be accounted for in probabilistic inference. In strictly asexual populations, the covariances have a particularly simple block-like structure imposed by the phylogenetic tree. Ho and Ane showed how this block-like structure can be exploited to efficiently invert covariance matrices and fit linear models on trees in linear time. In this short note, I use these methods to estimate evolutionary rates and to find the root of the tree that optimizes the time-divergence relationship. The algorithm is implemented in TreeTime and can be used to estimate evolutionary rates and their confidence intervals with computational cost scaling linearly in the number of tips.