Closed form modeling of evolutionary rates by exponential Brownian functionals (Q893813)

In this article, the authors present closed form integral expressions for the probability distribution of the pathwise average of evolutionary rates, assuming a bridged geometric Brownian model with diffusion parameter \(\sigma\) governing the time-dependent fluctuation of the rate itself. The average substitution rate along a branch of a phylogeny whose length corresponds to \(T\) calendar times is then obtained by integrating over the Brownian trajectories conditional on the rates at times \(0\) and \(T\). Using these expressions, the validity of the gamma approximation of the average rate distribution is assessed using a distribution fit based on closed form expression of the first two moments. When \(\sigma^2 T>4\), corresponding to stronger rate fluctuations and/or longer periods, the gamma approximation is no longer accurate. In such situations, the approximating distribution of the average rate along a branch is gamma with a small shape parameter, i.e., with a large proportion of small average rates and few very large values, which is far from the actual density function. As a result, the distribution of the transition probabilities between character states (nucleotide, amino acids or codons) in large time does not match the expected limiting distribution, which is clearly not satisfactory. However, authors' experience with the analysis of real data sets suggests that \(\sigma^2 T\) is generally smaller than 4, so the gamma approximation is likely to perform well in the practice.