Online Phylogenetics using Parsimony Produces Slightly Better Trees and is Dramatically More Efficient for Large SARS-CoV-2 Phylogenies than de novo and Maximum-Likelihood Approaches

Abstract

Phylogenetics has been foundational to SARS-CoV-2 research and public health policy, assisting in genomic surveillance, contact tracing, and assessing emergence and spread of new variants. However, phylogenetic analyses of SARS-CoV-2 have often relied on tools designed for de novo phylogenetic inference, in which all data are collected before any analysis is performed and the phylogeny is inferred once from scratch. SARS-CoV-2 datasets do not fit this mould. There are currently over 5 million sequenced SARS-CoV-2 genomes in public databases, with tens of thousands of new genomes added every day. Continuous data collection, combined with the public health relevance of SARS-CoV-2, invites an “online” approach to phylogenetics, in which new samples are added to existing phylogenetic trees every day. The extremely dense sampling of SARS-CoV-2 genomes also invites a comparison between Likelihood and Parsimony approaches to phylogenetic inference. Maximum Likelihood (ML) methods are more accurate when there are multiple changes at a single site on a single branch, but this accuracy comes at a large computational cost, and the dense sampling of SARS-CoV-2 genomes means that these instances will be extremely rare. Therefore, it may be that approaches based on Maximum Parsimony (MP) are sufficiently accurate for reconstructing phylogenies of SARS-CoV-2, and their simplicity means that they can be applied to much larger datasets. Here, we evaluate the performance of de novo and online phylogenetic approaches, and ML and MP frameworks, for inferring large and dense SARS-CoV-2 phylogenies. Overall, we find that online phylogenetics produces similar phylogenetic trees to de novo analyses for SARS-CoV-2, and that MP optimizations produce more accurate SARS-CoV-2 phylogenies than do ML optimizations. Since MP is thousands of times faster than presently available implementations of ML and online phylogenetics is faster than de novo, we therefore propose that, in the context of comprehensive genomic epidemiology of SARS-CoV-2, MP online phylogenetics approaches should be favored.

matOptimize: a parallel tree optimization method enables online phylogenetics for SARS-CoV-2

“…0.51% higher than the starting tree. This is consistent with our previous analyses ( Thornlow et al , 2021 ) in which we observed that parsimony and likelihood scores are strongly correlated for SARS-CoV-2 phylogenies. We also observed that in some cases, the PangoLEARN lineage assignments ( O’Toole et al , 2021 ) appeared a little less dispersed through the tree ( Supplementary Fig.…”

“…In contrast, TNT incurs a long startup delay followed by a slow improvement of parsimony. matOptimize is therefore especially well-suited for SARS-CoV-2 online phylogenetics ( Thornlow et al , 2021 ).…”

“…Online phylogenetics is a potentially transformative solution for the massive computational challenges imposed by continual genome sequence collection during the pandemic ( Gill et al , 2020 ; Thornlow et al , 2021 ). It involves incorporating new sequences as they become available onto an existing phylogeny and periodically refining the updated phylogeny.…”

“…The recent development of UShER ( Turakhia et al , 2021b ), which uses stepwise addition of new sequences onto an existing phylogenetic tree, helped overcome an important computational barrier for the maintenance of a comprehensive SARS-CoV-2 phylogenetic tree ( McBroome et al , 2021 ). For large SARS-CoV-2 phylogenies, UShER’s maximum parsimony (MP) approach has been demonstrated to perform well, even when compared to the far more computationally expensive maximum likelihood approaches available currently ( Thornlow et al , 2021 ). However, since UShER is based on the greedy approach of stepwise addition of new sequences, it sometimes places new sequences suboptimally in the tree ( Turakhia et al , 2021b ; Takahashi and Nei, 2000 ).…”

Taxonium, a web-based tool for exploring large phylogenetic trees

“…UShER takes a starting tree, built with iqtree or a similar approach, and incrementally adds sequences by maximum parsimony. For densely sampled sequencing efforts, as in the SARS-CoV-2 pandemic, such an approach still yields tree topologies of very high quality ( Thornlow et al, 2022 ). To turn this distance tree into a time tree, by estimating a time associated with each node in the tree, we recently developed Chronumental ( Sanderson, 2021 ) which uses stochastic gradient descent to efficiently construct chronologies from very large trees, which was not possible with previous approaches.…”