Table 1.

Orthology inference methods and databases used in this study

Tool/Dataset	Prediction type	Description and notes
Ancestral Panther http://ancestralgenomes.org	Database	Ancestral genomes dataset contains reconstructed ancestral genomes based on gene family trees from the PANTHER database, from which HMM profiles were built.
Broccoli https://github.com/rderelle/Broccoli	De novo prediction	K-mer preclustering to simplify proteomes, followed by a similarity search (DIAMOND) and phylogenetic analysis (FastTree2). Orthologous groups are inferred using a machine learning algorithm, LPA. Extremely fast when run on a large dataset.
EggNOG (DIAMOND and hmmer http://eggnog5.embl.de/#/app/home	Database	Manually curated sequence sets ran with (1) seed ortholog assignments (DIAMOND) and (2) HMM profile searches (hmmer).
Orthofinder (DIAMOND and BLAST) https://github.com/davidemms/OrthoFinder	De novo prediction	Uses both (1) DIAMOND or (2) BLAST as an aligner. Has a sequence length and phylogenetic distance normalized bit-score cut-off between pairs of genes, which function as edge weights in the orthogroup graph. Clustering of genes is done with the MCL method.
SonicParanoid http://iwasakilab.bs.s.u-tokyo.ac.jp/sonicparanoid/	De novo prediction	Uses MMseqs2 as an aligner. The algorithm of InParanoid is used as a backbone, with changes to the core algorithm that reduce the execution time and increase the usability of the tool. Relies on cumulative alignment score of groups and avoids using thresholds based on confidence score between pairs of genes. Clustering of genes is done with the MCL method.
SwiftOrtho https://github.com/Rinoahu/SwiftOrtho	De novo prediction	Taking the same approach as OrthoMCL for normalized bit-score cut-off between pairs of genes, which function as the edge weights in the graph. Clustering of genes is done with the MCL method. SwiftOrtho is optimized for speed and memory usage when applied to large-scale data.