hmdb-database

name	hmdb-database
description	Parse HMDB (Human Metabolome Database) local XML for metabolite info, chemical properties, biological context, disease links, spectra, and cross-DB mapping. No REST API — uses ~6 GB XML download. Use drugbank-database-access for drugs; pubchem-compound-search for live lookups.
license	CC-BY-4.0

HMDB Database — Local XML Access

Overview

Query the Human Metabolome Database (HMDB, 220,000+ metabolite entries) by parsing locally downloaded XML with Python's ElementTree. Covers metabolite lookup, chemical properties, biological context (pathways, enzymes, biofluids), disease/biomarker associations, spectral data for metabolite identification, and cross-database ID mapping to KEGG, PubChem, ChEBI, and DrugBank.

When to Use

• Looking up metabolite information (description, chemical class, cellular location) by HMDB ID or name
• Retrieving chemical properties (molecular weight, formula, SMILES, InChI, logP, PSA) for metabolomics analysis
• Finding pathway associations and enzyme links for a set of metabolites
• Identifying biofluid/tissue locations of metabolites (blood, urine, CSF, saliva)
• Querying disease associations and normal/abnormal concentration ranges for biomarker discovery
• Extracting NMR or MS spectral peak lists for metabolite identification
• Mapping HMDB IDs to KEGG, PubChem, ChEBI, DrugBank, or other databases
• For drug-specific data (interactions, targets, pharmacology) use drugbank-database-access instead
• For live compound property queries without downloading use pubchem-compound-search instead

Prerequisites

• HMDB XML download: Register at https://hmdb.ca/downloads — download hmdb_metabolites.xml.zip (~6 GB uncompressed)
• Python packages: lxml (faster XPath) or standard xml.etree.ElementTree, pandas
• No public REST API: HMDB has no programmatic REST API. All access is via local XML parsing or the web interface
• R package (optional): hmdbQuery on CRAN provides some query wrappers but is limited and outdated
• Rate limits: N/A for local XML parsing. The web interface has no documented rate limits but is not intended for scraping

pip install lxml pandas

Quick Start

import xml.etree.ElementTree as ET

NS = {'hmdb': 'http://www.hmdb.ca'}

tree = ET.parse('hmdb_metabolites.xml')  # 60-120s for full XML
root = tree.getroot()

# Build lookup index (HMDB ID + lowercase name -> element)
metabolite_index = {}
for met in root.findall('hmdb:metabolite', NS):
    accession = met.find('hmdb:accession', NS)
    name = met.find('hmdb:name', NS)
    if accession is not None and accession.text:
        metabolite_index[accession.text] = met
    if name is not None and name.text:
        metabolite_index[name.text.lower()] = met

def find_metabolite(query):
    """Find metabolite by HMDB ID or name (case-insensitive)."""
    return metabolite_index.get(query) or metabolite_index.get(query.lower())

met = find_metabolite('HMDB0000122')  # Glucose
name = met.find('hmdb:name', NS).text
formula = met.find('hmdb:chemical_formula', NS).text
print(f"{name}: {formula}")
# Glucose: C6H12O6

Core API

1. XML Setup and Metabolite Lookup

import xml.etree.ElementTree as ET

NS = {'hmdb': 'http://www.hmdb.ca'}
tree = ET.parse('hmdb_metabolites.xml')
root = tree.getroot()
print(f"Total metabolite entries: {len(root.findall('hmdb:metabolite', NS))}")
# Total metabolite entries: ~220000+

For memory-constrained environments, use iterparse:

met_names = {}
for event, elem in ET.iterparse('hmdb_metabolites.xml', events=('end',)):
    if elem.tag == '{http://www.hmdb.ca}metabolite':
        acc = elem.find('{http://www.hmdb.ca}accession')
        name = elem.find('{http://www.hmdb.ca}name')
        if acc is not None and name is not None and acc.text and name.text:
            met_names[acc.text] = name.text
        elem.clear()
print(f"Parsed {len(met_names)} metabolites via iterparse")

2. Chemical Properties

def get_chemical_properties(met_element):
    """Extract chemical properties from a metabolite entry."""
    def txt(path):
        el = met_element.find(path, NS)
        return el.text if el is not None and el.text else None
    # Fields: accession, name, chemical_formula, average_molecular_weight,
    # monisotopic_molecular_weight, smiles, inchi, inchikey, state, iupac_name
    return {tag: txt(f'hmdb:{tag}') for tag in [
        'accession', 'name', 'chemical_formula', 'average_molecular_weight',
        'monisotopic_molecular_weight', 'smiles', 'inchi', 'inchikey', 'state']}

props = get_chemical_properties(find_metabolite('HMDB0000158'))  # L-Tyrosine
print(f"{props['name']}: MW={props['average_molecular_weight']}, SMILES={props['smiles']}")

# Extract taxonomy / chemical classification (ClassyFire ontology)
def get_classification(met_element):
    tax = met_element.find('hmdb:taxonomy', NS)
    if tax is None:
        return {}
    def txt(tag):
        el = tax.find(f'hmdb:{tag}', NS)
        return el.text if el is not None and el.text else None
    return {'kingdom': txt('kingdom'), 'super_class': txt('super_class'),
            'class': txt('class'), 'sub_class': txt('sub_class'),
            'direct_parent': txt('direct_parent')}

print(get_classification(find_metabolite('HMDB0000122')))
# {'kingdom': 'Organic compounds', 'super_class': 'Organooxygen compounds', ...}

3. Biological Context

def get_pathways(met_element):
    """Extract metabolic pathway associations."""
    pathways = []
    for pw in met_element.findall('hmdb:pathways/hmdb:pathway', NS):
        name = pw.find('hmdb:name', NS)
        smpdb_id = pw.find('hmdb:smpdb_id', NS)
        kegg_id = pw.find('hmdb:kegg_map_id', NS)
        pathways.append({
            'name': name.text if name is not None else None,
            'smpdb_id': smpdb_id.text if smpdb_id is not None else None,
            'kegg_map_id': kegg_id.text if kegg_id is not None else None,
        })
    return pathways

for pw in get_pathways(find_metabolite('HMDB0000122'))[:5]:
    print(f"  {pw['smpdb_id']}: {pw['name']}")

def get_biolocations(met_element):
    """Extract biofluid, tissue, and cellular locations."""
    bp = 'hmdb:biological_properties/hmdb:'
    def texts(path):
        return [el.text for el in met_element.findall(path, NS) if el.text]
    return {'biofluids': texts(f'{bp}biospecimen_locations/hmdb:biospecimen'),
            'tissues': texts(f'{bp}tissue_locations/hmdb:tissue'),
            'cellular': texts(f'{bp}cellular_locations/hmdb:cellular')}

locs = get_biolocations(find_metabolite('HMDB0000122'))
print(f"Glucose biofluids: {locs['biofluids']}")

def get_enzymes(met_element):
    """Extract associated enzymes/proteins with UniProt IDs."""
    return [{'name': (p.find('hmdb:name', NS).text
                      if p.find('hmdb:name', NS) is not None else None),
             'uniprot_id': (p.find('hmdb:uniprot_id', NS).text
                            if p.find('hmdb:uniprot_id', NS) is not None else None),
             'gene_name': (p.find('hmdb:gene_name', NS).text
                           if p.find('hmdb:gene_name', NS) is not None else None)}
            for p in met_element.findall('hmdb:protein_associations/hmdb:protein', NS)]

enz = get_enzymes(find_metabolite('HMDB0000122'))
print(f"Glucose-associated proteins: {len(enz)}")
for e in enz[:3]:
    print(f"  {e['gene_name']} ({e['uniprot_id']}): {e['name']}")

4. Disease and Biomarker Queries

def get_diseases(met_element):
    """Extract disease associations with OMIM IDs and PubMed references."""
    diseases = []
    for d in met_element.findall('hmdb:diseases/hmdb:disease', NS):
        name = d.find('hmdb:name', NS)
        omim = d.find('hmdb:omim_id', NS)
        pmids = [r.find('hmdb:pubmed_id', NS).text
                 for r in d.findall('hmdb:references/hmdb:reference', NS)
                 if r.find('hmdb:pubmed_id', NS) is not None and r.find('hmdb:pubmed_id', NS).text]
        diseases.append({'name': name.text if name is not None else None,
                         'omim_id': omim.text if omim is not None else None,
                         'pubmed_ids': pmids})
    return diseases

diseases = get_diseases(find_metabolite('HMDB0000122'))
print(f"Glucose disease associations: {len(diseases)}")
for d in diseases[:3]:
    print(f"  {d['name']} (OMIM: )")

def get_concentrations(met_element, biospecimen='Blood'):
    """Extract normal/abnormal concentration data filtered by biospecimen."""
    result = {'normal': [], 'abnormal': []}
    for ctype, key in [('normal_concentrations', 'normal'),
                       ('abnormal_concentrations', 'abnormal')]:
        for c in met_element.findall(f'hmdb:{ctype}/hmdb:concentration', NS):
            bio = c.find('hmdb:biospecimen', NS)
            if bio is None or bio.text != biospecimen:
                continue
            def txt(tag):
                el = c.find(f'hmdb:{tag}', NS)
                return el.text if el is not None else None
            result[key].append({'value': txt('concentration_value'),
                                'units': txt('concentration_units'),
                                'condition': txt('subject_condition')})
    return result

conc = get_concentrations(find_metabolite('HMDB0000122'), 'Blood')
print(f"Glucose blood:  normal,  abnormal")

5. Spectral Data

def get_ms_spectra(met_element):
    """Extract MS/MS spectral peak lists (m/z + intensity)."""
    spectra = []
    for spec in met_element.findall('hmdb:spectra/hmdb:spectrum', NS):
        stype = spec.find('hmdb:type', NS)
        if stype is None or 'MS' not in (stype.text or ''):
            continue
        peaks = [{'mz': float(p.find('hmdb:mass_charge', NS).text),
                  'intensity': float(p.find('hmdb:intensity', NS).text or 0)}
                 for p in spec.findall('hmdb:ms_ms_peaks/hmdb:ms_ms_peak', NS)
                 if p.find('hmdb:mass_charge', NS) is not None
                 and p.find('hmdb:mass_charge', NS).text]
        spectra.append({'type': stype.text, 'num_peaks': len(peaks), 'peaks': peaks})
    return spectra

spectra = get_ms_spectra(find_metabolite('HMDB0000122'))
print(f"Glucose MS spectra: {len(spectra)}")

def get_nmr_spectra(met_element):
    """Extract NMR spectral peak lists (1H, 13C). Same pattern as MS above."""
    spectra = []
    for spec in met_element.findall('hmdb:spectra/hmdb:spectrum', NS):
        stype = spec.find('hmdb:type', NS)
        if stype is None or 'NMR' not in (stype.text or ''):
            continue
        nucleus = spec.find('hmdb:nucleus', NS)
        shifts = [float(p.find('hmdb:chemical_shift', NS).text)
                  for p in spec.findall('hmdb:nmr_one_d_peaks/hmdb:nmr_one_d_peak', NS)
                  if p.find('hmdb:chemical_shift', NS) is not None
                  and p.find('hmdb:chemical_shift', NS).text]
        spectra.append({'type': stype.text,
                        'nucleus': nucleus.text if nucleus is not None else None,
                        'num_peaks': len(shifts), 'chemical_shifts': shifts})
    return spectra

nmr = get_nmr_spectra(find_metabolite('HMDB0000122'))
print(f"Glucose NMR spectra: ")

6. Cross-Database Mapping

def get_external_ids(met_element):
    """Extract cross-database identifiers (KEGG, PubChem, ChEBI, DrugBank, CAS, etc.)."""
    fields = {'kegg_id': 'KEGG', 'pubchem_compound_id': 'PubChem', 'chebi_id': 'ChEBI',
              'drugbank_id': 'DrugBank', 'chemspider_id': 'ChemSpider',
              'cas_registry_number': 'CAS', 'biocyc_id': 'BioCyc', 'pdb_id': 'PDB',
              'foodb_id': 'FooDB', 'metlin_id': 'METLIN'}
    ids = {}
    for tag, label in fields.items():
        el = met_element.find(f'hmdb:{tag}', NS)
        if el is not None and el.text:
            ids[label] = el.text
    return ids

ids = get_external_ids(find_metabolite('HMDB0000122'))
print(f"Glucose cross-refs: {ids}")
# {'KEGG': 'C00031', 'PubChem': '5793', 'ChEBI': '17234', ...}

# Build cross-reference table for a metabolite list
import pandas as pd

queries = ['HMDB0000122', 'HMDB0000158', 'HMDB0000167', 'HMDB0000148']
rows = []
for q in queries:
    m = find_metabolite(q)
    if m is None: continue
    ids = get_external_ids(m)
    rows.append({'name': m.find('hmdb:name', NS).text, 'hmdb_id': q,
                 'kegg': ids.get('KEGG'), 'pubchem': ids.get('PubChem'),
                 'chebi': ids.get('ChEBI')})
print(pd.DataFrame(rows).to_string(index=False))

Key Concepts

HMDB XML Entry Structure

Section	XPath	Content
Identity	hmdb:accession, hmdb:name, hmdb:iupac_name	Primary identifiers
Chemical	hmdb:chemical_formula, hmdb:smiles, hmdb:inchi	Structure descriptors
Properties	hmdb:average_molecular_weight, hmdb:state	Physical properties
Taxonomy	hmdb:taxonomy/hmdb:kingdom, class, etc.	Chemical classification
Biological	hmdb:biological_properties	Biofluids, tissues, cellular locations
Pathways	hmdb:pathways/hmdb:pathway	SMPDB + KEGG pathway links
Enzymes	hmdb:protein_associations/hmdb:protein	Associated proteins/enzymes
Diseases	hmdb:diseases/hmdb:disease	Disease associations + OMIM IDs
Concentrations	hmdb:normal_concentrations, hmdb:abnormal_concentrations	Biomarker reference ranges
Spectra	hmdb:spectra/hmdb:spectrum	NMR, MS/MS peak lists
External IDs	hmdb:kegg_id, hmdb:pubchem_compound_id, etc.	Cross-database identifiers
Ontology	hmdb:ontology	Physiological/disposition/process roles

Data Field Completeness

Not all entries have all fields populated. Coverage varies by metabolite class:

Field Category	Approximate Coverage	Notes
Accession, name, formula	~100%	Always present
SMILES, InChI, MW	~90%	Missing for some lipids and complex metabolites
Taxonomy/classification	~85%	Chemical ontology from ClassyFire
Biofluid locations	~60%	Best for common human metabolites
Pathways	~40%	Curated SMPDB + KEGG links
Protein associations	~35%	Enzyme-metabolite relationships
Disease associations	~25%	Primarily for biomarker metabolites
Normal concentrations	~20%	Reference ranges for clinical metabolites
MS/MS spectra	~15%	Experimental spectral libraries
NMR spectra	~10%	1H and 13C chemical shift data

File Format Comparison

Format	File	Size	Best For
Full XML	hmdb_metabolites.xml	~6 GB	Complete data access (all fields)
SDF	structures.sdf	~200 MB	Chemical structures + basic properties
CSV	Various exports	~50-500 MB	Tabular data (properties, concentrations)
FASTA	hmdb_proteins.fasta	~50 MB	Protein sequence lookups

Common Workflows

Workflow 1: Metabolite Identification from Mass Spec

Goal: Match an observed m/z value to candidate metabolites using molecular weight. Assumes root, NS from Quick Start.

import pandas as pd

observed_mz = 180.063  # [M+H]+ for glucose
adduct_mass = 1.00728  # H+ adduct
target_mw = observed_mz - adduct_mass
tolerance_da = 0.01

candidates = []
for met in root.findall('hmdb:metabolite', NS):
    mw_el = met.find('hmdb:monisotopic_molecular_weight', NS)
    if mw_el is None or not mw_el.text:
        continue
    mw = float(mw_el.text)
    if abs(mw - target_mw) <= tolerance_da:
        candidates.append({
            'hmdb_id': met.find('hmdb:accession', NS).text,
            'name': met.find('hmdb:name', NS).text,
            'mw': mw, 'delta_da': abs(mw - target_mw),
            'formula': (met.find('hmdb:chemical_formula', NS).text
                        if met.find('hmdb:chemical_formula', NS) is not None else None),
        })

df = pd.DataFrame(candidates).sort_values('delta_da')
print(f"Candidates within {tolerance_da} Da of {target_mw:.3f}: {len(df)}")
print(df.head(10).to_string(index=False))

Workflow 2: Biomarker Discovery for a Disease

Goal: Find all metabolites associated with a disease and their concentration changes. Assumes root, NS from Quick Start.

import pandas as pd

disease_query = 'diabetes'
biomarkers = []
for met in root.findall('hmdb:metabolite', NS):
    for d in met.findall('hmdb:diseases/hmdb:disease', NS):
        dname = d.find('hmdb:name', NS)
        if dname is None or not dname.text:
            continue
        if disease_query.lower() not in dname.text.lower():
            continue
        abnormal = [c for c in met.findall(
            'hmdb:abnormal_concentrations/hmdb:concentration', NS)
            if c.find('hmdb:subject_condition', NS) is not None
            and disease_query.lower() in
            (c.find('hmdb:subject_condition', NS).text or '').lower()]
        biomarkers.append({
            'hmdb_id': met.find('hmdb:accession', NS).text,
            'metabolite': met.find('hmdb:name', NS).text,
            'disease': dname.text,
            'abnormal_measurements': len(abnormal),
        })

df = pd.DataFrame(biomarkers).drop_duplicates(['hmdb_id', 'disease'])
print(f"Metabolites linked to '{disease_query}': ")
(df.sort_values(, ascending=).head().to_string(index=))

Workflow 3: Pathway Enrichment from Metabolite List

Goal: Given a list of identified metabolites, find over-represented pathways. Assumes metabolite_index from Quick Start.

from collections import Counter
import pandas as pd

hit_ids = ['HMDB0000122', 'HMDB0000158', 'HMDB0000167',
           'HMDB0000148', 'HMDB0000064', 'HMDB0000161']

hit_pathways = Counter()
for hid in hit_ids:
    met = metabolite_index.get(hid)
    if met is None:
        continue
    for pw in met.findall('hmdb:pathways/hmdb:pathway', NS):
        pw_name = pw.find('hmdb:name', NS)
        if pw_name is not None and pw_name.text:
            hit_pathways[pw_name.text] += 1

enriched = [(pw, count) for pw, count in hit_pathways.most_common() if count >= 2]
df = pd.DataFrame(enriched, columns=['Pathway', 'Hit_Count'])
print(f"Pathways with 2+ hits from {len(hit_ids)} metabolites:")
print(df.to_string(index=False))

Key Parameters

Parameter	Function/Endpoint	Default	Description
NS (namespace dict)	All XPath queries	{'hmdb': 'http://www.hmdb.ca'}	Required for all find/findall calls
tolerance_da	MW matching	0.01	Mass tolerance in Daltons for metabolite ID
biospecimen	get_concentrations()	'Blood'	Filter: Blood, Urine, Cerebrospinal Fluid (CSF), Saliva, etc.
events	ET.iterparse()	('end',)	Parse events; use ('end',) to fire on closing tags
adduct mass	MS identification	varies	1.00728 [M+H]+, 22.9892 [M+Na]+, -1.00728 [M-H]-
target_type	Spectral queries	'MS' or 'NMR'	Filter spectra by type string

Best Practices

1. Build an in-memory index on startup: Parse once (60-120s), build dict by HMDB ID + lowercase name. Never re-parse inside a loop
2. Always pass the namespace dict: Every find()/findall() needs NS = {'hmdb': 'http://www.hmdb.ca'}. Omitting it returns empty results
3. Use iterparse for memory constraints: Full XML requires ~8-12 GB RAM. Use elem.clear() in iterparse to process incrementally
4. Guard against None and empty text: Many optional fields are absent. Always check el is not None and el.text before accessing
5. Use monoisotopic weight for MS matching: monisotopic_molecular_weight is correct for mass spec; average_molecular_weight for other calculations
6. Pre-filter by chemical class for large searches: Use taxonomy/classification to narrow searches before iterating all 220K+ entries

Common Recipes

Recipe: Export All Metabolite Properties to CSV

When to use: Create a flat table of all metabolites with key properties.

import pandas as pd
records = []
for met in root.findall('hmdb:metabolite', NS):
    def txt(p):
        el = met.find(p, NS)
        return el.text if el is not None and el.text else None
    records.append({'hmdb_id': txt('hmdb:accession'), 'name': txt('hmdb:name'),
                    'formula': txt('hmdb:chemical_formula'),
                    'avg_mw': txt('hmdb:average_molecular_weight'),
                    'smiles': txt('hmdb:smiles')})
pd.DataFrame(records).to_csv('hmdb_properties.csv', index=False)
print(f"Exported {len(records)} metabolites")

Recipe: Find Metabolites by Biofluid

When to use: Get all metabolites detected in a specific biofluid (e.g., urine for clinical screening).

biofluid = 'Urine'
urine_mets = []
for met in root.findall('hmdb:metabolite', NS):
    for bf in met.findall(
            'hmdb:biological_properties/hmdb:biospecimen_locations/hmdb:biospecimen', NS):
        if bf.text and bf.text == biofluid:
            urine_mets.append({
                'hmdb_id': met.find('hmdb:accession', NS).text,
                'name': met.find('hmdb:name', NS).text,
            })
            break
print(f"Metabolites in {biofluid}: {len(urine_mets)}")

Recipe: Chemical Class Distribution

When to use: Summarize metabolite chemical classes for a hit list or the full database.

from collections import Counter
class_counts = Counter()
for met in root.findall('hmdb:metabolite', NS):
    tax = met.find('hmdb:taxonomy', NS)
    if tax is not None:
        sc = tax.find('hmdb:super_class', NS)
        if sc is not None and sc.text:
            class_counts[sc.text] += 1
print(dict(class_counts.most_common(10)))

Troubleshooting

Problem	Cause	Solution
find() returns None for known elements	Missing XML namespace	Always pass NS = {'hmdb': 'http://www.hmdb.ca'}
MemoryError parsing full XML	~8-12 GB needed in memory	Use ET.iterparse() with elem.clear()
Slow startup (>120s)	Parsing 6 GB XML	Parse once, build index dict; avoid re-parsing
Metabolite not found by name	Case sensitivity or synonym	Normalize to lowercase; try HMDB ID directly
Empty spectra for a metabolite	Not all entries have spectra (~10-15% coverage)	Check coverage table; use METLIN or MassBank for more spectra
Missing concentration data	Limited to well-studied clinical metabolites (~20%)	Cross-reference with MetaboAnalyst or literature
Duplicate entries for same compound	Secondary accessions (HMDB00XXXXX vs HMDB0000XXXX)	Use accession (primary), not secondary_accessions
ET.iterparse missing data	Premature elem.clear()	Only clear after extracting all needed fields from the element

Bundled Resources

Self-contained entry. The original reference file hmdb_data_fields.md (268 lines, field catalog with XML element names and descriptions) is consolidated into the Key Concepts "HMDB XML Entry Structure" table and the "Data Field Completeness" table above. The field catalog's per-element XML tag names are demonstrated in Core API code blocks. Omitted from original: web interface descriptions (not programmatic).

Related Skills

• drugbank-database-access -- Drug-specific data (interactions, targets, pharmacology) with similar local XML parsing pattern
• pubchem-compound-search -- Live compound property lookups without downloading; PubChemPy REST API
• kegg-database -- Pathway and compound database with REST API for cross-referencing HMDB pathway hits
• matchms-spectral-matching -- Spectral similarity matching against reference libraries; complementary to HMDB spectral data
• pyopenms-mass-spectrometry -- Full LC-MS/MS processing pipeline; use HMDB for metabolite identification step

References

• HMDB website and downloads: https://hmdb.ca/
• Wishart DS et al. (2022). HMDB 5.0: the Human Metabolome Database for 2022. Nucleic Acids Res. 50(D1):D801-D816. https://doi.org/10.1093/nar/gkab1062
• HMDB data download page: https://hmdb.ca/downloads
• MetaboAnalyst (complementary tool): https://www.metaboanalyst.ca/