Phylogenetic Analysis of SARS-CoV-2 Data Is Difficult

Abstract

Numerous studies covering some aspects of SARS-CoV-2 data analyses are being published on a daily basis, including a regularly updated phylogeny on nextstrain.org. Here, we review the difficulties of inferring reliable phylogenies by example of a data snapshot comprising a quality-filtered subset of 8, 736 out of all 16, 453 virus sequences available on May 5, 2020 from gisaid.org. We find that it is difficult to infer a reliable phylogeny on these data due to the large number of sequences in conjunction with the low number of mutations. We further find that rooting the inferred phylogeny with some degree of confidence either via the bat and pangolin outgroups or by applying novel computational methods on the ingroup phylogeny does not appear to be credible. Finally, an automatic classification of the current sequences into sub-classes using the mPTP tool for molecular species delimitation is also, as might be expected, not possible, as the sequences are too closely related. We conclude that, although the application of phylogenetic methods to disentangle the evolution and spread of COVID-19 provides some insight, results of phylogenetic analyses, in particular those conducted under the default settings of current phylogenetic inference tools, as well as downstream analyses on the inferred phylogenies, should be considered and interpreted with extreme caution.

SARS-CoV-2 within-host diversity and transmission

“…We followed the recommendations of Morel et al . ( 42 ), in which 100 separate maximum likelihood phylogenies were generated using RAxML-NG ( 66 ) and the GTR+G substitution model, such that each reconstruction used a different random starting parsimony tree. The final phylogeny was then obtained from this set using majority rule.…”

“…We sought to gain a better understanding of SARS-CoV-2 evolution and to determine whether iSNVs could be used to help resolve phylogenies and transmission clusters. For the 1390 genomes in our study, we constructed a phylogeny using the robust procedure outlined by ( 42 ) ( Fig. 5A ).…”

“…The phylogeny was constructed by maximum likelihood according to the robust procedure outlined by Morel et al . ( 42 ).…”

SARS-CoV-2 convergent evolution cannot be reliably inferred from phylogenetic analyses

“…The lack of statistical support for homoplasy for the vast majority of mutations results from the general lack of phylogenetic signal in SARS-CoV-2 sequences, as has been noted previously (Morel et al, 2020). In addition, several other factors may play an important role.…”

“…Lack of phylogenetic signal greatly increases the size of the plausible tree set, whilst model misspecification alters the trees included in this set and their relative likelihood values. In this case, the tree with the highest likelihood (the 'best-fitting' tree) is selected, more or less at random, from an immense set of plausible trees with nearly equal support (Morel et al, 2020), which makes it extremely unlikely that this "best" tree represents the true evolutionary history. It is difficult to conclude that some feature of the evolutionary dynamics, such as homoplasy, is supported by the phylogeny if the set of plausible trees includes trees that do and do not support that feature.…”