SATURN: six-species OrthoFinder with a choanoflagellate outgroup

Analysis
Mnemiopsis leidyi
Comparative genomics
HPC & computing
Published

October 20, 2026

Analysis·Project Mnemiopsis Phagocytes·Tools OrthoFinder 3.1.0 DIAMOND MAFFT FastTree Ensembl 113 NCBI RefSeq Python LSF
ImportantSummary
  • A new orthology reference covers all six SATURN species:
    • Mnemiopsis, oyster, fly, Ciona, zebrafish, human;
    • plus the choanoflagellate Salpingoeca rosetta as an outgroup, to root the tree.
  • The outgroup fixed the missing Metazoa level. The 4-species run of Oct 15 wrote no root-level hierarchical orthogroups. Here the root is Choanozoa and node N1 = Metazoa, so the Metazoa-level groups exist (N1.tsv).
  • 137,235 genes (one protein per gene), 85.6% in 13,282 orthogroups:
    • 3,255 orthogroups contain all seven species;
    • 1,260 are single-copy in every species (Table 4).
  • Gene placement follows divergence and genome history (Figure 1):
    • human and zebrafish have 93% of genes in orthogroups;
    • Mnemiopsis, fly and Ciona 78–80%;
    • oyster has 31.5% of its genes in oyster-only orthogroups, the most of any species. Its branch carries 16,601 gene duplications.
  • One-to-one orthologs: human–zebrafish 9,208 pairs; Mnemiopsis pairs 3,062–4,384. Oyster–Ciona (5,607) has more than Ciona–zebrafish (4,055), because zebrafish genes are often duplicated.
  • Two gene-level reference tables for the SATURN benchmarks:
    • orthogroups: 29,292 groups;
    • Metazoa-level groups: 30,275 groups.

Key terms

Table 1: Terms used in this entry.
Term Meaning here
Orthogroup (OG) All genes descended from one gene in the last common ancestor of the species analysed (orthologs plus in-paralogs).
Hierarchical orthogroup (HOG) An orthogroup defined at one node of the species tree using gene trees. A HOG at the Metazoa node holds the genes descended from one gene in the last common ancestor of animals. Duplications older than that node split groups.
Outgroup A species outside the group of interest, used to root the tree. Here S. rosetta, a choanoflagellate (the closest living relatives of animals).
Species-specific orthogroup An orthogroup containing genes from only one species, typically a lineage-specific family or expansion.
Unassigned gene A gene placed in no orthogroup: no detectable relative among the other proteomes.
One-to-one orthologs A gene pair where each species has exactly one copy descended from the ancestral gene.
Gene duplication (per node) A duplication event OrthoFinder places on a branch of the species tree from the gene trees.

1. Why a new run

  • Fly and human are new. The eggNOG annotation tables and the Oct 15 OrthoFinder run cover only the original four species.
  • Metazoa-level HOGs were missing.
    • The Oct 15 run used a fixed tree with Mnemiopsis at the base, so the Metazoa node was the root.
    • OrthoFinder 3.1 wrote no N0.tsv (root-level HOGs). That left only orthogroups, which can lump paralogs from very old duplications.
    • Adding a non-animal outgroup puts the Metazoa node one step below the root.
  • S. rosetta has a well-annotated RefSeq proteome (~11.6k genes) and is a standard choanoflagellate outgroup.

2. Inputs

One protein per gene: the longest. Gene IDs use the same protein-to-gene bridges as the SATURN embeddings (06a). Internal stop codons are replaced by X.

Table 2: OrthoFinder input (one longest protein per gene).
Species Proteome Genes Notes
Mnemiopsis leidyi genome-v3 gene models 15,545 as on Oct 15
Crassostrea gigas (oyster) NCBI RefSeq 31,371 as on Oct 15
Ciona robusta Ghost KY21 18,788 as on Oct 15
Danio rerio (zebrafish) Ensembl 113 GRCz11 25,400 as on Oct 15 (unnamed alternate-haplotype copies dropped)
Drosophila melanogaster (fly) Ensembl 113 BDGP6.46 13,986
Homo sapiens (human) Ensembl 113 GRCh38 20,527 primary assembly only: 11,512 proteins (3,337 genes) on alternate-haplotype (HSCHR*) and patch (HG*) scaffolds dropped; 98% duplicate a primary-assembly gene and would otherwise look like paralogs. Unplaced primary scaffolds (KI*, GL*) kept
Salpingoeca rosetta (outgroup) NCBI RefSeq GCF_000188695.1 11,618 11,725 proteins, all linked to a gene via GFF locus_tag
Total 137,235

3. The run

  • Command: OrthoFinder 3.1.0, -M msa -A mafft -T fasttree -t 16 -a 16.
  • Species tree: (Sros,(Mlei,((Cgig,Dmel),(Crob,(Drer,Hsap))))), fixed with -s.
    • It assumes ctenophores branch first among animals. With a single ctenophore and no sponge, the sponge-first hypothesis gives the same Metazoa node.
  • Compute: Pegasus parallel queue (one 16-core node).
    • 4.3 h (15,617 s), 56 CPU-hours.
    • Peak memory 41.5 GB.
    • The 4-species run took 70 min and 15.9 GB.
  • Node labels assigned by OrthoFinder:
Table 3: Species-tree nodes.
Node Clade Species below it HOG file
N0 Choanozoa (root) all 7 not written (OrthoFinder 3.1 does not output the root)
N1 Metazoa 6 SATURN species N1.tsv
N2 Bilateria oyster, fly, Ciona, zebrafish, human N2.tsv
N3 Protostomia oyster, fly N3.tsv
N4 Deuterostomia Ciona, zebrafish, human N4.tsv
N5 Vertebrata zebrafish, human N5.tsv

4. Results

4.1 Overall

Table 4: OrthoFinder summary statistics.
Statistic 6 species + outgroup (this run) 4 species (Oct 15)
Genes 137,235 91,104
Genes in orthogroups 117,535 (85.6%) 75,861 (83.3%)
Unassigned genes 19,700 (14.4%) 15,243 (16.7%)
Orthogroups 13,282 (mean 8.8, median 6 genes) 12,254 (mean 6.2, median 4)
Species-specific orthogroups 3,543, holding 22,876 genes (16.7%) 3,561, holding 22,576 genes (24.8%)
Orthogroups with all species 3,255 4,987
… single-copy in every species 1,260 2,577
G50 / O50 (assigned genes) 11 genes / 2,085 orthogroups 5 genes / 3,263 orthogroups

Orthogroups present in all species and single-copy orthogroups both fall with more species. Each extra species is another chance for a family to be lost or duplicated.

Table 5: Orthogroups by number of species.
Species in the orthogroup 1 2 3 4 5 6 7
Orthogroups 3,543 1,814 792 878 1,220 1,780 3,255

4.2 Per species

Table 6: Per-species statistics (genes = one protein per gene).
Mnemiopsis oyster fly Ciona zebrafish human S. rosetta
Genes 15,545 31,371 13,986 18,788 25,400 20,527 11,618
% in orthogroups 78.5 90.8 79.8 77.6 93.3 93.3 71.0
% in species-specific orthogroups 20.1 31.5 14.2 15.6 7.1 6.9 14.9
% unassigned 21.5 9.2 20.2 22.4 6.7 6.7 29.0
Orthogroups containing the species 6,308 9,156 6,194 7,468 9,014 8,951 5,533
Duplications on its own branch 4,681 16,601 3,625 5,106 8,537 4,425 2,400
Figure 1: Left: share of each species’ genes in multi-species orthogroups, species-specific orthogroups, or unassigned. Right: one-to-one ortholog pairs between each pair of species.
  • Vertebrates are the best connected: 86% of human and zebrafish genes are in multi-species orthogroups. That reflects their gene content and their mutual closeness: the panel has two vertebrates and one of each other lineage.
  • Mnemiopsis, fly and Ciona have about a fifth of their genes unassigned. For Mnemiopsis this caps any orthology-based score: genes without a detectable relative cannot be judged right or wrong by an orthology reference.
  • Oyster stands out. 31.5% of its genes are in oyster-only orthogroups, and its branch carries more duplications (16,601) than any other. That fits the known lineage-specific gene family expansions in oysters.
  • Zebrafish has the second-highest branch duplication count (8,537), consistent with the teleost genome duplication.

4.3 One-to-one orthologs

Table 7: One-to-one ortholog pairs between the six SATURN species.
Mnemiopsis oyster fly Ciona zebrafish human
Mnemiopsis 4,245 3,546 4,384 3,062 3,292
oyster 4,549 5,607 4,085 4,333
fly 4,531 3,368 3,273
Ciona 4,055 4,303
zebrafish 9,208
  • Human–zebrafish is by far the richest pair (9,208), which makes it the natural positive control for cross-species cell matching.
  • Every Mnemiopsis pair has 3,000–4,400 one-to-one orthologs: fewest with the vertebrates, most with Ciona and oyster.
  • Oyster–Ciona (5,607) exceeds Ciona–zebrafish (4,055), although Ciona and zebrafish are the closer relatives. Duplications on the zebrafish lineage turn many one-to-one relationships into one-to-many; the same was seen with four species on Oct 15.

4.4 Duplications on internal branches

Table 8: Gene duplications placed on internal branches of the species tree.
Node Duplications (all) Duplications (≥50% support)
N1 Metazoa 1,047 591
N2 Bilateria 1,340 680
N3 Protostomia 93 93
N4 Deuterostomia 200 171
N5 Vertebrata 4,251 4,251

The vertebrate stem carries most internal-branch duplications, as expected from the two vertebrate whole-genome duplications. Only 591 well-supported duplications sit at the Metazoa node, so most Metazoa-level HOGs match their orthogroup.

4.5 Reference tables for the SATURN benchmarks

06c turns the output into gene-level tables (species, gene, protein, group) with the outgroup removed. It finds the Metazoa node automatically from the species tree: the node whose descendants are exactly the six species.

Table 9: Gene-level reference tables (six SATURN species).
Table Groups (incl. singletons) Genes in groups of ≥2 Genes in groups with all 6 species
Orthogroups (ALL6_gene_orthofinder_OG.csv) 29,292 109,141 (87% of 125,617) 46,272 (37%)
Metazoa HOGs (ALL6_gene_orthofinder_N1HOG.csv) 30,275 108,798 41,426 (33%)

The HOG table splits orthogroups at duplications older than Metazoa, so it is the stricter reference: fewer groups span all six species. Both will be used to score the 6-species SATURN macrogene runs.

Caveats

  • Fixed species tree, ctenophores first. HOG levels depend on it. With one ctenophore and no sponge the Metazoa node is the same under either hypothesis.
  • Annotation quality differs between species. Mnemiopsis and Ciona gene models are less complete than human or fly, which raises their unassigned share. Gene-model completeness (e.g. BUSCO) would make the comparison fairer.
  • One protein per gene (longest). Other isoforms can carry different domains; this matches standard OrthoFinder practice.
  • Peak memory (41.5 GB) exceeded the 32 GB requested. It fitted only because the node had 62.8 GB. Request ≥48 GB for runs this size.

Next steps

Scripts

All in pipelines/scripts/saturn_embedding_comparison/.

Table 10: Scripts used in this entry.
Script What it does
06a0_download_outgroup.sh downloads the S. rosetta proteome and GFF from NCBI RefSeq (transfers queue)
06a_orthofinder_prep.py longest protein per gene, gene bridges, human primary-assembly filter, species tree (--species)
06b6_orthofinder_6sp.sh OrthoFinder 3.1.0 job (micromamba orthofinder_env, FastTree preflight)
06c_orthofinder_to_reference.py gene-level OG and Metazoa-HOG tables; outgroup removed; Metazoa node from the species tree 
cd /scratch/dark_genes/SATURN_Mnemi
bsub < scripts/06a0_download_outgroup.sh
python3 scripts/06a_orthofinder_prep.py --species Mlei,Cgig,Crob,Drer,Dmel,Hsap,Sros --out_dir 06_orthofinder_6sp
bsub -q parallel < scripts/06b6_orthofinder_6sp.sh          # job 603391 (moved from normal with bmod)
python scripts/06c_orthofinder_to_reference.py --of_dir 06_orthofinder_6sp --saturn_run ""

References

  • Emms D.M., Kelly S. (2019) OrthoFinder: phylogenetic orthology inference for comparative genomics. Genome Biology 20:238.
  • Emms D.M., Liu Y., Belcher L., Holmes J., Kelly S. (2025) OrthoFinder 3 (bioRxiv; as cited by OrthoFinder 3.1.0).
  • Emms D.M., Kelly S. (2017) STRIDE: species tree root inference from gene duplication events. Molecular Biology and Evolution 34:3267–3278.
  • Fairclough S.R. et al. (2013) Premetazoan genome evolution and the regulation of cell differentiation in the choanoflagellate Salpingoeca rosetta. Genome Biology 14:R15.
  • Zhang G. et al. (2012) The oyster genome reveals stress adaptation and complexity of shell formation. Nature 490:49–54.
Back to top