Bio Ortholog Inference

Version Compatibility

Reference examples tested with: requests 2.31+, pandas 2.2+; OrthoDB v12 API, Ensembl REST (Ensembl release 112+), OMA REST API, eggNOG 6.0+, PANTHER v18+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show requests pandas
• API surface: confirm endpoint URLs and JSON schema match the current API docs

If endpoints return 404 or unexpected JSON, check release notes for the resource; schema migrations happen with each major version (Ensembl release is the biggest moving target).

Ortholog Inference (Database Access)

"What is the X ortholog of gene Y?" -> Many ortholog resources have already done the inference at scale. Pulling their answers is faster and often more reliable than re-computing. This skill is the database-access view: how to query the major orthology resources programmatically, what their confidence semantics mean, and when their disagreements matter.

For de novo orthology inference (running OrthoFinder, SonicParanoid, OMA standalone on local proteomes), see comparative-genomics/ortholog-inference — that's a much deeper treatment of the computational side.

This skill is about pulling answers from:

• OrthoDB v12 — broadest coverage (1700+ species), levels from species-specific to deep

• Ensembl Compara — vertebrate-focused, tree-reconciled, confidence scores

• OMA browser — high precision, HOG (Hierarchical Orthologous Group) framework

• eggNOG 6.0 — pre-computed functional groups, deepest functional annotation

• PANTHER — protein family + ortholog calls with experimentally validated curation

• KEGG Orthology (KO) — pathway-centric orthologous functional units

• HomoloGene — deprecated since 2014, but data still queryable for legacy comparison

• Python: requests.get() against REST endpoints; pandas for parsing

• CLI: curl against the same endpoints; OrthoDB also has bulk downloads

Required Setup

import requests
import pandas as pd
import time

No API keys required for any of these resources (as of 2026), but rate limits apply — see per-resource notes below.

Decision matrix: which resource for which question?

Per-resource API reference

OrthoDB v12

Base URL: https://data.orthodb.org/v12/

Key endpoints (all GET, JSON returned):

• /search?query=<symbol>&species=<NCBI_taxid> — find ortholog groups by gene symbol
• /orthologs?id=<og_id>&species=<taxid> — get orthologs of a group at a specific level
• /group?id=<og_id> — full group info (sequences, evidence)
• /tab?query=<og_id> — tab-separated bulk dump

Levels are NCBI taxonomy IDs (e.g. 9606 = human, 40674 = Mammalia, 7742 = Vertebrata).

def orthodb_search(symbol, species_taxid=9606):
    r = requests.get('https://data.orthodb.org/v12/search',
                     params={'query': symbol, 'species': species_taxid})
    r.raise_for_status()
    return r.json()['data']  # list of orthogroup IDs

Ensembl Compara (via Ensembl REST)

Base URL: https://rest.ensembl.org/. JSON: Accept: application/json. Rate limit: 15 req/sec, 55,000 req/hour. Respect Retry-After on 429.

Key endpoints:

• /homology/symbol/<species>/<symbol> — orthologs of a gene by symbol
• /homology/id/<ensembl_gene_id> — orthologs of a gene by Ensembl ID
• /lookup/symbol/<species>/<symbol> — resolve symbol to Ensembl ID first
• Add ?type=orthologues to filter to orthologs only (drop paralogs)
• Add ?target_species=<species> to filter to one target species

def ensembl_orthologs(symbol, species='human', target=None):
    url = f'https://rest.ensembl.org/homology/symbol/{species}/{symbol}'
    params = {'type': 'orthologues'}
    if target:
        params['target_species'] = target
    r = requests.get(url, params=params, headers={'Accept': 'application/json'})
    r.raise_for_status()
    homologies = r.json()['data'][0]['homologies']
    return [{
        'target_species': h['target']['species'],
        'target_id': h['target']['id'],
        'type': h['type'],  # ortholog_one2one / one2many / many2many / within_species_paralog
        'confidence': h.get('confidence'),  # 0/1; some calls lack this field
        'identity_target': h['target'].get('perc_id'),
        'identity_query': h['source'].get('perc_id'),
    } for h in homologies]

OMA REST API

Base URL: https://omabrowser.org/api/. JSON returned. No rate-limit doc but be polite.

Key endpoints:

• /protein/<id>/orthologs/ — orthologs of a protein (UniProt or OMA ID)
• /hog/<hog_id>/ — Hierarchical Orthologous Group info
• /genome/<species_code>/ — list all genomes; species codes are 5-letter (e.g. HUMAN, MOUSE)

def oma_orthologs(uniprot_acc):
    r = requests.get(f'https://omabrowser.org/api/protein/{uniprot_acc}/orthologs/')
    r.raise_for_status()
    return r.json()  # list of ortholog dicts with omaid, canonicalid, taxonId

eggNOG (5/6)

Base URL: http://eggnog6.embl.de/api/ (web API; lighter than running eggNOG-mapper). Most heavy lifting still uses eggNOG-mapper locally (Cantalapiedra et al. 2021 Mol Biol Evol 38:5825) — for batch protein-set annotation, mapper > API.

For ad hoc lookup: search the eggNOG web interface for an orthogroup ID, then download the member set.

KEGG Orthology (KO)

Base URL: https://rest.kegg.jp/. Returns plain text TSV by default (NOT JSON).

def kegg_ko_for_gene(species_code, gene):
    '''KEGG species codes: hsa=human, mmu=mouse, dme=fly, etc.'''
    r = requests.get(f'https://rest.kegg.jp/link/ko/{species_code}:{gene}')
    r.raise_for_status()
    return [line.split('\t')[1].replace('ko:', '') for line in r.text.strip().split('\n') if line]


def kegg_orthologs(ko_id):
    r = requests.get(f'https://rest.kegg.jp/link/genes/{ko_id}')
    return [line.split('\t')[1] for line in r.text.strip().split('\n') if line]

KEGG license: commercial use requires a paid license; academic use is free for web/API.

PANTHER

Base URL: http://pantherdb.org/services/oai/pantherdb/ (note the unusual base). Has a curated, smaller scope than OrthoDB but with experimental evidence.

HomoloGene (deprecated)

https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=homologene&id=<id>&rettype=xml — still works; data frozen since 2014. Useful only for backward-compatibility with old pipelines.

Confidence-level semantics

Each resource defines "confidence" differently. They are NOT directly comparable.

Don't average or compare confidence across resources. Use within-resource cutoffs; for cross-resource comparison, intersect call sets.

The orthology conjecture (and why resources disagree)

The orthology conjecture (Tatusov 1997; rigorously evaluated by Studer & Robinson-Rechavi 2009 Trends Genet 25:210; Altenhoff et al. 2012 PLoS Comput Biol 8:e1002514) — orthologs are more likely than paralogs to share function — is supported but weakly. Sub- and neo-functionalization mean a paralog can become the functional equivalent.

This is also why resources disagree. Different algorithms emphasize different evidence:

• OMA is strict (RBH + verification + HOG inference) — higher precision, lower recall.
• Ensembl Compara is tree-reconciled — best for vertebrates with deep Compara curation.
• OrthoDB uses broader hierarchical clustering — broader coverage, more ambiguous calls.
• eggNOG uses pre-computed orthogroups at fixed taxonomic levels — fast but coarser.

For high-stakes calls (publication, drug target choice), intersect at least two resources and inspect disagreements.

Code patterns

Get the human ortholog of a mouse gene (Ensembl Compara, 1:1 only)

Goal: Pull Compara's high-confidence 1:1 ortholog of a single mouse gene in human.

Approach: REST query with type filter; assert 1:1; record confidence.

Reference (Ensembl REST, release 112+):

import requests
import time


def compara_one2one(symbol, source='mouse', target='human'):
    url = f'https://rest.ensembl.org/homology/symbol/{source}/{symbol}'
    r = requests.get(url, params={'type': 'orthologues', 'target_species': target},
                     headers={'Accept': 'application/json'})
    if r.status_code == 429:
        time.sleep(int(r.headers.get('Retry-After', '5')))
        return compara_one2one(symbol, source, target)
    r.raise_for_status()
    hits = r.json()['data'][0]['homologies']
    one2one = [h for h in hits if h['type'] == 'ortholog_one2one']
    if not one2one:
        return None
    h = one2one[0]
    return {
        'source_id': h['source']['id'],
        'target_id': h['target']['id'],
        'confidence': h.get('confidence'),
        'pid_target': h['target'].get('perc_id'),
    }

Cross-resource agreement (Ensembl + OMA + OrthoDB)

Goal: Find orthologs agreed on by multiple resources to flag high-confidence calls.

Approach: Query each resource; intersect target IDs after normalizing to a common namespace (UniProt or NCBI Gene).

def cross_resource_orthologs(symbol):
    '''Return target-species ortholog calls from multiple resources for cross-validation.'''
    ensembl = ensembl_orthologs(symbol, species='human')
    # (OMA/OrthoDB lookups omitted for brevity -- need namespace conversion via UniProt ID Mapping)
    return {'ensembl': ensembl}

Batch ortholog table for >100 genes

Goal: Build a wide table of orthologs across N species for a gene list.

Approach: Loop with rate limit; cache responses; respect Retry-After.

Reference (requests 2.31+):

def batch_ensembl_orthologs(symbols, source='human', target_species=None, sleep=0.07):
    '''sleep=0.07 keeps under the 15-req/sec ceiling with margin.'''
    rows = []
    for sym in symbols:
        try:
            orthologs = ensembl_orthologs(sym, species=source, target=target_species)
            for o in orthologs:
                rows.append({'source_symbol': sym, **o})
        except requests.HTTPError as e:
            if e.response.status_code == 429:
                wait = int(e.response.headers.get('Retry-After', '10'))
                time.sleep(wait)
            else:
                rows.append({'source_symbol': sym, 'error': str(e)})
        time.sleep(sleep)
    return pd.DataFrame(rows)


df = batch_ensembl_orthologs(['BRCA1', 'TP53', 'MYC'], target_species='mouse')
print(df[df['type'] == 'ortholog_one2one'][['source_symbol', 'target_id', 'confidence']])