Historian: accurate reconstruction of ancestral sequences and evolutionary rates

Abstract

Reconstruction of ancestral sequence histories, and estimation of parameters like indel rates, are improved by using explicit evolutionary models and summing over uncertain alignments. The previous best tool for this purpose (according to simulation benchmarks) was ProtPal, but this tool was too slow for practical use.Historian combines an efficient reimplementation of the ProtPal algorithm with performance-improving heuristics from other alignment tools. Simulation results on fidelity of rate estimation via ancestral reconstruction, along with evaluations on the structurally-informed alignment dataset BAliBase 3.0, recommend Historian over other alignment tools for evolutionary applications. Historian

BAli-Phy version 3: model-based co-estimation of alignment and phylogeny

“…Model-based inference then requires a probabilistic model of insertion-deletion events on each branch of the tree, as well as phylogenetic substitution likelihoods to score substitutions during alignment construction. A variety of programs implement model-based alignment, including Historian (Holmes, 2017), PRANK (Loeytynoja and Goldman, 2005), ProPIP (Maiolo et al, 2018). Model-based inference of alignments can increase alignment accuracy by up to 3-fold on simulated data sets (Redelings, 2014) when compared to the popular MSA construction tools MUSCLE (Edgar, 2004) and MAFFT (Katoh, et al, 2005).…”

CherryML: Scalable Maximum Likelihood Estimation of Phylogenetic Models

“…We used the EM implementation provided by the Historian package [19]. We ran Historian’s command with the following arguments:…”

“…To estimate the time that it would take for traditional methods such as zeroth-order optimization or EM to estimate a 400 × 400 co-evolution rate matrix, we performed the following extrapolation. From Figure 1b we observe that it takes EM as implemented by Historian [19] around 30 CPU-hours to learn a single-site model on 1,024 families with 128 sequences each. Since the runtime of traditional methods scales linearly in the dataset size, this implies that learning a single-site model on all 15,051 families with approximately 1,024 sequences each would take on the order of CPU-hours.…”

“…By using simulated data, we benchmarked the computational runtime and statistical efficiency of the CherryML method for the aforementioned LG model, and compared it with the performance of EM, for which we used the implementation in the Historian package [19]. The results are summarized in Figure 1b,c, which shows that CherryML is more than a thousand times faster than EM when run on 1,024 families with 128 sequences each, taking 19 seconds as compared to almost 30 hours.…”