Coral proteome functional annotation

Functional annotation

Comparative genomics

HPC & computing

Annotate every gene model in a coral proteome with orthologs, domains, signal peptides, transmembrane topology and a functional class, merged into one master table per species. Eight LSF jobs per species.

Published

June 5, 2026

Modified

October 16, 2026

Pipeline·Tools eggNOG-mapper v2 OrthoFinder 3.1.0 InterProScan 5.78-109.0 DIAMOND 2.2.1 SignalP 6.0 DeepTMHMM 1.0 HMMER 3.4

Overview

Coral genomes are full of genes with no name. This pipeline gathers every kind of evidence available for each predicted protein and merges it into a single master annotation table per species (53 columns, including the protein sequence). The table plugs straight into single-cell annotation, comparative genomics and cell–cell communication (CellChat) analyses.

Every gene gets four independent labels:

Axis	Values	Meaning
Confidence tier	Gold > Silver > Bronze > Informative > Unclassified	How strong the homology evidence is
Annotation status	Light / Partial / Dark	Named function / domain only / no evidence (“dark genes”), following Stephens et al. 2026
Protein group	24 functional groups	For grouping genes in Seurat and CellChat
CellChat role	Ligand / Receptor / Receptor_candidate / Neither	From signal peptide and transmembrane predictions

Code and full documentation

All scripts live in the public GitHub repository kevinhwong1/coral-func-annotation-pipeline. It also holds finished annotation tables for 10 cnidarian species, the column schema and the classification logic. This page is the quick guide.

How it runs

flowchart TB
  P[(Proteome FASTA + GFF3)] --> A
  P --> J5
  subgraph A [Phase 1: all run in parallel]
    J1[01 eggNOG-mapper<br/>orthology, GO, KEGG]
    J2[02 OrthoFinder<br/>vs human, mouse, Nvec]
    J3[03 InterProScan<br/>domains]
    J4[04 DIAMOND RBH<br/>vs SwissProt]
    J6[06 HMMER<br/>179 custom HMMs]
    J7[07 DIAMOND<br/>curated coral refs]
  end
  J5[05 SignalP 6] --> T[05b DeepTMHMM<br/>2,000-seq chunks]
  T --> M{{Phase 2: merge DeepTMHMM<br/>chunks, manual}}
  A --> F[08 merge_annotate.py]
  M --> F
  F --> O[(Master annotation table + gene lists)]

Job	Tool	Purpose	Queue	Walltime
01	eggNOG-mapper v2	Orthology, GO, KEGG, COG	bigmem	12 h
02	OrthoFinder 3.1.0	Orthologs vs human, mouse and N. vectensis	bigmem	120 h
03	InterProScan 5.78-109.0	Domain annotation (17 databases)	bigmem	48 h
04	DIAMOND 2.2.1	Reciprocal best hits vs SwissProt	bigmem	8 h
05	SignalP 6.0	Signal peptides	general	20 h
05b	DeepTMHMM 1.0	Transmembrane topology, in 2,000-sequence chunks	general	12 h per chunk
06	HMMER 3.4	Tier 3 custom HMM profiles	general	6 h
07	DIAMOND 2.2.1	Tier 4 curated coral BLAST	general	1 h
08	`08_merge_annotate.py`	Merge all layers → master table	general	2 h

Run it

1. One-time setup

You need:

Anaconda in your home directory and Java 11+ (InterProScan).
Academic licenses for SignalP 6.0 and DeepTMHMM, requested from DTU Health Tech. Put the license tarballs in your home directory.
The N. vectensis proteome (NV2g.20240221.protein.fa) from SimrBase, used as the OrthoFinder outgroup.

git clone https://github.com/kevinhwong1/coral-func-annotation-pipeline.git
cd coral-func-annotation-pipeline

Edit the top-level paths in pipeline/run_config.sh for your account and cluster:

export NETHOME="<fill in: your home directory>"
export BASE="<fill in: scratch folder for databases and runs>"
export EMAIL="<fill in: email for LSF notifications>"
export LSF_PROJECT="<fill in: LSF project/allocation>"

Then build the conda environments and download all databases. This takes 2–4 h; the eggNOG database alone is ~39 GB.

bash pipeline/00_setup_annotation_env.sh

2. Submit a species

source ~/anaconda3/etc/profile.d/conda.sh
conda activate annotation_env

# a species already configured in run_config.sh
bash pipeline/submit_pipeline.sh Pdam

# any new species: pass the paths as flags (no config edits needed)
bash pipeline/submit_pipeline.sh <SPECIES_CODE> \
    --proteome        <fill in: /path/to/proteome.fasta> \
    --gff3            <fill in: /path/to/annotation.gff3> \
    --genome_version  "<fill in: e.g. MySpecies_v1.0>" \
    --proteome_source "<fill in: source URL>"

This writes every job script to <BASE>/runs/<SPECIES>_<date>/logs/ and submits them with LSF dependencies.

3. Merge the DeepTMHMM chunks, then the final merge runs

DeepTMHMM chunks can fail individually, so merging them is a manual check:

RUN_DIR=<fill in: <BASE>/runs/<SPECIES>_<date>>
TMHMM_DIR="${RUN_DIR}/05_tmhmm"

# every chunk should say OK; resubmit missing ones with: bsub < ${RUN_DIR}/logs/job_05b_chunk_<NN>.bsub
for d in "${TMHMM_DIR}"/results/chunk_*/; do
  [[ -f "${d}/TMRs.gff3" ]] && echo "$(basename $d): OK" || echo "$(basename $d): MISSING"
done

echo "##gff-version 3" > "${TMHMM_DIR}/TMRs.gff3"
for d in "${TMHMM_DIR}"/results/chunk_*/; do grep -v "^##gff-version" "${d}/TMRs.gff3" >> "${TMHMM_DIR}/TMRs.gff3"; done

bsub < "${RUN_DIR}/logs/job_08_merge.bsub"

Monitor with bjobs | grep <SPECIES> and bpeek <JID>. Full restart and manual-merge instructions are in the repository’s pipeline README.

Outputs

In <RUN_DIR>/08_final/:

File	What it is
`<SPECIES>_master_annotation.tsv`	The table to use: one row per gene, 53 columns (tier, status, protein group, best human ortholog, GO, KEGG, Pfam/InterPro, signal peptide, TM helices, CellChat role, sequence…)
`gene_lists/`	Light / Partial / Dark gene lists, CellChat ligands and receptors, one list per protein group
`<SPECIES>_annotation_summary.md`	Tier and group counts for the run
`<SPECIES>_load_gene_lists.R`	Loads every gene list into R in one line

Join it to Seurat cluster markers by gene ID. Check that the ID formats match first: CellBender/GeneExt feature IDs often differ from the proteome IDs (e.g. pdam_00025168 vs pdam_00000427-RA), so normalize them before joining.

Results so far

Species	Genes	Gold	Silver	Bronze	Light	Partial	Dark
Galaxea fascicularis	22,418	31.2%	23.0%	10.7%	71.6%	9.9%	18.4%
Pocillopora damicornis	26,077	27.1%	21.6%	10.1%	64.3%	9.3%	26.4%
Acropora cervicornis	28,059	24.4%	18.5%	8.3%	62.0%	10.0%	28.0%
Nematostella vectensis	32,495	21.5%	24.6%	12.0%	65.7%	11.1%	23.2%

Six more species (A. millepora, A. muricata, O. faveolata, O. patagonica, S. pistillata, Xenia sp.) are in the repository.

Known issues

The full list (~30 fixes) is in the repository’s pipeline README. These are the ones most likely to catch you:

DeepTMHMM: compute node only, 12 h per chunk, run from its install folder

It loads a 2.6 GB protein language model and is killed on login nodes. 2,000 sequences take ~9 h, so a 6 h walltime fails. It must cd into its install directory to find its model files, and it crashes if the output folder already exists.

Pin numpy<2.0 and fair-esm==0.4.0

numpy 2.x breaks SignalP and fair-esm; fair-esm 2.0 breaks DeepTMHMM’s predict.py.

SignalP fails on long FASTA headers

Strip headers to the first word before running (needed for P. damicornis). Otherwise SignalP throws an OSError.

NCBI genomes: 0 reciprocal best hits and broken coordinates

NCBI BLAST databases prefix subject IDs with gb|accession|, so reverse hits don’t match forward queries; the RBH parser strips the wrapper. NCBI GFF3 mRNA IDs (gnl|WGS|…) also differ from the protein accessions, so a gene-to-protein table is needed (pass it with --gene_to_protein). Affects A. cervicornis, O. faveolata and A. muricata.

Conda activation inside job scripts

Use source ~/anaconda3/etc/profile.d/conda.sh (not bin/activate) and hard-coded environment names inside BSUB heredocs. Otherwise conda activate can read the species name as the environment name.

Keep OrthoFinder in its own environment

orthofinder_env (OrthoFinder, MAFFT, FastTree) must stay separate from annotation_env.

How it connects

Multi-species orthology (OrthoFinder): the combined orthogroups used to transfer cell-type markers between species.
scRNA-seq QC and clustering: master tables are joined to cluster markers during annotation.
Earlier notebook entries: KofamScan workflow, KofamScan on Galaxea.

References

Stephens TJ et al. 2026. Widespread dark gene evolution in stony corals. Genome Biol Evol 18:evag072. https://doi.org/10.1093/gbe/evag072
Cantalapiedra CP et al. 2021 (eggNOG-mapper v2); Emms & Kelly 2019 (OrthoFinder); Jones et al. 2014 (InterProScan); Buchfink et al. 2021 (DIAMOND); Teufel et al. 2022 (SignalP 6.0); Hallgren et al. 2022 (DeepTMHMM); Eddy 2011 (HMMER).

Changelog

Date	Change
2026-10-05	Notebook page added, summarizing the GitHub repository
2026-07	v4: species-agnostic gene-ID columns, NCBI RBH/GFF3 fixes; 10 species complete
2026-06	First complete runs (Galaxea, Pocillopora, Acropora cervicornis)