Benchmarking orthology methods using phylogenetic patterns defined at the base of Eukaryotes

Abstract

Insights into the evolution of ancestral complexes and pathways are generally achieved through careful and time-intensive manual analysis often using phylogenetic profiles of the constituent proteins. This manual analysis limits the possibility of including more protein-complex components, repeating the analyses for updated genome sets or expanding the analyses to larger scales. Automated orthology inference should allow such large-scale analyses, but substantial differences between orthologous groups generated by different approaches are observed. We evaluate orthology methods for their ability to recapitulate a number of observations that have been made with regard to genome evolution in eukaryotes. Specifically, we investigate phylogenetic profile similarity (co-occurrence of complexes), the last eukaryotic common ancestor’s gene content, pervasiveness of gene loss and the overlap with manually determined orthologous groups. Moreover, we compare the inferred orthologies to each other. We find that most orthology methods reconstruct a large last eukaryotic common ancestor, with substantial gene loss, and can predict interacting proteins reasonably well when applying phylogenetic co-occurrence. At the same time, derived orthologous groups show imperfect overlap with manually curated orthologous groups. There is no strong indication of which orthology method performs better than another on individual or all of these aspects. Counterintuitively, despite the orthology methods behaving similarly regarding large-scale evaluation, the obtained orthologous groups differ vastly from one another. Availability and implementation The data and code underlying this article are available in github and/or upon reasonable request to the corresponding author: https://github.com/ESDeutekom/ComparingOrthologies.

Coexpression reveals conserved mechanisms of transcriptional cell identity

“…Our coexpression conservation analysis is made possible by the use of OrthoDB to define orthologous genes. However, orthology prediction is an active area of research in genomics 23,43 , and relying on a single algorithm has known limitations 44–46 . How would our results hold up if we had used a different reference?…”

“…A primary application of orthology prediction is to infer shared function across species, but benchmarks recurrently find a precision-recall trade-off across algorithms 46,59 , with little evidence that any one approach outperforms another 43 . By incorporating functional information directly, coexpression conservation scores may improve sequence-based inference.…”

The spread of the first introns in proto-eukaryotic paralogs

“…To reconstruct ancestral intron positions we used a diverse set of 167 eukaryotic (predicted) proteomes, as compiled for a previous study (28). In that study, these proteins had been assigned to the different eukaryotic eggNOG families (euNOGs) (29) using hidden Markov model profile searches (28). Sverdlov et al .…”

“…The used species phylogeny can be found in Supplementary Figure 1 of Deutekom et al . (2021) (28). Because the position of the eukaryotic root remains under debate (36), we also used a tree with an unresolved root between Diaphoretickes, Amorphea, Metamonada and Discoba.…”

“…To reconstruct ancestral intron positions we used a diverse set of 167 eukaryotic (predicted) proteomes, as compiled for a previous study (28). In that study, these proteins had been assigned to the different eukaryotic eggNOG families (euNOGs) (29) using hidden Markov model profile searches (28). Sverdlov et al (2007) (17) used the homologous clusters of eukaryotic orthologous groups (KOGs) and candidate orthologous groups (TWOGs) from Makarova et al (2005) (14).…”