chembl-database-bioactivity

ChEMBL Database — Bioactivity Queries

Why no SDK? The chembl_webresource_client package is convenient sugar over a public, no-auth REST/JSON API at https://www.ebi.ac.uk/chembl/api/data/. When the SDK is unavailable, every operation can be reproduced with plain requests and URL parameters. This SKILL.md uses the REST path throughout so the code runs in any environment with requests installed. Django-style filter syntax (field__icontains=…, field__lte=…, field__range=a,b) works as URL query parameters.

Overview

ChEMBL is EMBL-EBI's bioactive molecule database: 2M+ compounds, 19M+ bioactivity measurements (IC50, Ki, EC50, Kd, …), 13K+ targets. The REST API at https://www.ebi.ac.uk/chembl/api/data/ returns JSON (append .json) or XML/YAML, requires no authentication, and supports Django-style query filters via URL parameters plus cursor-style pagination via page_meta.next.

When to Use

• Finding compounds by name, ChEMBL ID, or physicochemical properties
• Querying bioactivity data (IC50, Ki, EC50) for specific targets
• Performing similarity or substructure searches using SMILES
• Retrieving drug mechanisms of action and clinical indications
• Identifying inhibitors, agonists, or bioactive molecules for a target
• Analyzing structure-activity relationships (SAR) across compound series
• Filtering molecules by Lipinski rule-of-5 or other drug-likeness criteria
• For general cheminformatics (SMILES manipulation, fingerprints, descriptors) use rdkit-cheminformatics instead
• For an alternative compound database (NIH, broader coverage) use pubchem-compound-search

Prerequisites

• Python packages: requests (only requirement). Optional: pandas for tabular analysis.
• No API key required: ChEMBL is freely accessible.
• Rate limits: No published hard limit. The infrastructure is shared — add time.sleep(0.2-0.5) between requests in batch loops; back off on HTTP 429.

pip install requests
# Optional, for DataFrame work:
pip install pandas

Quick Start

import requests

BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Retrieve a molecule by ChEMBL ID
r = requests.get(f"{BASE}/molecule/CHEMBL25.json", timeout=15)
r.raise_for_status()
aspirin = r.json()
print(f"{aspirin['pref_name']}: MW={aspirin['molecule_properties']['mw_freebase']}")
# ASPIRIN: MW=180.16

# Search targets by full name (acronyms like 'EGFR' don't match pref_name — use full term)
r = requests.get(
    f"{BASE}/target.json",
    params={"pref_name__icontains": "epidermal growth factor receptor",
            "target_type": "SINGLE PROTEIN", "limit": 5},
    timeout=15,
)
targets = r.json()["targets"]
print(f"EGFR-like targets: {len(targets)}, first={targets[0]['target_chembl_id']}")

# Potent bioactivities: EGFR (CHEMBL203) IC50 <= 100 nM
r = requests.get(
    f"{BASE}/activity.json",
    params={"target_chembl_id": "CHEMBL203",
            "standard_type": "IC50",
            "standard_value__lte": 100,
            "standard_units": "nM",
            "limit": 5},
    timeout=30,
)
data = r.json()
print(f"EGFR IC50 ≤ 100 nM records: {data['page_meta']['total_count']}")

Key Concepts

Filter Operators (Django-style, as URL parameters)

The SDK's field__operator=value syntax maps 1:1 to URL query parameters. Use & to combine filters.

Operator	URL pattern	Example URL fragment
__exact	field=value	target_type=SINGLE+PROTEIN
__iexact	field__iexact=value	pref_name__iexact=aspirin
__contains / __icontains	field__icontains=value	pref_name__icontains=kinase
__startswith / __endswith	field__startswith=Epi	pref_name__endswith=nib
__gt / __gte / __lt / __lte	field__lte=100	standard_value__lte=100
__range	field__range=lo,hi	molecule_properties__mw_freebase__range=300,500
__in	field__in=a,b,c	standard_type__in=IC50,Ki,Kd
__isnull	field__isnull=False (Python False/True strings)	pchembl_value__isnull=False
__regex	field__regex=…	pref_name__regex=^EGF.*kinase$
__search	field__search=…	description__search=apoptosis

When passed via requests.get(..., params={...}), the library handles URL encoding automatically (including the commas in __range and __in).

Core Endpoints

All endpoints accept .json, .xml, or .yaml suffix. JSON is the default below.

Endpoint URL	Returns	Key fields
/molecule/{chembl_id}.json	Compound by ID	pref_name, molecule_chembl_id, molecule_properties, molecule_structures
/molecule.json?<filters>	Compound search	paginated molecules[]
/target/{chembl_id}.json	Target by ID	pref_name, target_type, organism, target_components
/target.json?<filters>	Target search	paginated targets[]
/activity.json?<filters>	Bioactivity records	paginated activities[]
/assay.json?<filters>	Assay details	paginated assays[]
/drug.json?<filters>	Approved drug info	paginated drugs[]; supports /drug/{chembl_id}.json
/mechanism.json?<filters>	Mechanism of action	paginated mechanisms[]
/drug_indication.json?<filters>	Therapeutic indications	paginated drug_indications[]
/similarity/{smiles}/{tanimoto}.json	Tanimoto similarity (0–100)	paginated molecules[] with similarity field
/substructure/{smiles}.json	Substructure search	paginated molecules[]
/image/{chembl_id}.svg	SVG structure image	binary SVG (NOT JSON)
/molecule_form/{chembl_id}.json	Parent/salt forms	molecule_forms[]
/protein_class.json	Protein classification hierarchy	hierarchical browse
/document.json?<filters>	Literature source records	paginated documents[]

Response Shape

{
  "page_meta": {
    "limit": 20,
    "offset": 0,
    "total_count": 12145,
    "next": "/chembl/api/data/activity.json?...&offset=20",
    "previous": null
  },
  "activities": [ /* or molecules[], targets[], etc. */ ]
}

Walk page_meta.next (a relative URL — prefix with https://www.ebi.ac.uk) until it becomes null.

Molecular Properties

Properties accessible via molecule_properties on each record:

Field	Description
mw_freebase	Molecular weight (free base)
full_mwt	Full molecular weight (including salts)
alogp	Calculated LogP
hba	Hydrogen bond acceptors
hbd	Hydrogen bond donors
psa	Polar surface area
rtb	Rotatable bonds
num_ro5_violations	Lipinski rule-of-5 violations
ro3_pass	Rule of 3 compliance
cx_most_apka / cx_most_bpka	Most acidic / basic pKa

Target Information Fields

Field	Description
target_chembl_id	ChEMBL target identifier
pref_name	Preferred (full) target name — acronyms like "EGFR" do NOT match; use the spelled-out term
target_type	SINGLE PROTEIN, PROTEIN COMPLEX, ORGANISM, …
organism	Target organism (e.g., Homo sapiens)
tax_id	NCBI taxonomy ID
target_components[]	Components (UniProt accession, sequence, …)

Bioactivity Data Fields

Field	Description
standard_type	Activity type: IC50, Ki, Kd, EC50, …
standard_value	Numerical activity value
standard_units	Units: nM, uM, …
pchembl_value	Normalized -log10 activity (>6 = potent)
activity_comment	Activity annotations
data_validity_comment	Data quality flags (check before analysis)
potential_duplicate	Duplicate flag

Core API

1. Molecule Queries

import requests
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# By ChEMBL ID
r = requests.get(f"{BASE}/molecule/CHEMBL25.json", timeout=15)
aspirin = r.json()
print(f"{aspirin['pref_name']}: MW={aspirin['molecule_properties']['mw_freebase']}")

# By name (case-insensitive substring)
r = requests.get(f"{BASE}/molecule.json",
                 params={"pref_name__icontains": "imatinib", "limit": 5},
                 timeout=15)
for mol in r.json()["molecules"]:
    print(f"  {mol['molecule_chembl_id']}  {mol.get('pref_name')!r}")

# By Lipinski-compliant property ranges
r = requests.get(f"{BASE}/molecule.json",
                 params={"molecule_properties__mw_freebase__range": "300,500",
                         "molecule_properties__alogp__lte": 5,
                         "molecule_properties__hba__lte": 10,
                         "molecule_properties__hbd__lte": 5,
                         "limit": 3},
                 timeout=15)
print(f"Lipinski-compliant total: {r.json()['page_meta']['total_count']}")

2. Target Queries

import requests
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# By ChEMBL ID
r = requests.get(f"{BASE}/target/CHEMBL203.json", timeout=15)
egfr = r.json()
print(f"{egfr['pref_name']} ({egfr['organism']}) — type={egfr['target_type']}")

# Search by full name (NOT acronym) + type
r = requests.get(f"{BASE}/target.json",
                 params={"pref_name__icontains": "kinase",
                         "target_type": "SINGLE PROTEIN", "limit": 5},
                 timeout=15)
d = r.json()
print(f"Kinase SINGLE_PROTEIN targets: total={d['page_meta']['total_count']}")
for t in d["targets"][:5]:
    print(f"  {t['target_chembl_id']:12s} {t.get('pref_name')!r}  ({t['organism']})")

# By organism
r = requests.get(f"{BASE}/target.json",
                 params={"organism": "Homo sapiens", "limit": 3},
                 timeout=15)
print(f"Human targets: total={r.json()['page_meta']['total_count']}")

3. Bioactivity Data

import requests
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Potent inhibitors for a target (EGFR = CHEMBL203)
r = requests.get(f"{BASE}/activity.json",
                 params={"target_chembl_id": "CHEMBL203",
                         "standard_type": "IC50",
                         "standard_value__lte": 100,
                         "standard_units": "nM",
                         "limit": 5},
                 timeout=30)
data = r.json()
print(f"EGFR IC50≤100nM: total={data['page_meta']['total_count']}")
for act in data["activities"][:5]:
    print(f"  {act['molecule_chembl_id']:14s} IC50={act['standard_value']} nM "
          f"pChEMBL={act.get('pchembl_value')}")

# All pChEMBL-tagged activities for a compound
r = requests.get(f"{BASE}/activity.json",
                 params={"molecule_chembl_id": "CHEMBL25",
                         "pchembl_value__isnull": "False",
                         "limit": 5},
                 timeout=30)
print(f"Aspirin pChEMBL activities: total={r.json()['page_meta']['total_count']}")

# Multiple activity types (CHEMBL240 = D2 dopamine receptor)
r = requests.get(f"{BASE}/activity.json",
                 params={"target_chembl_id": "CHEMBL240",
                         "standard_type__in": "IC50,Ki,Kd",
                         "limit": 5},
                 timeout=30)
print(f"D2 receptor IC50/Ki/Kd: total={r.json()['page_meta']['total_count']}")

4. Structure-Based Search

import requests
from urllib.parse import quote
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Similarity search (Tanimoto ≥ 85%)
# Path-style endpoint: /similarity/{smiles}/{threshold}
# The SMILES MUST be URL-encoded (it contains '/', '(', ')' etc.)
aspirin_smiles = quote("CC(=O)Oc1ccccc1C(=O)O", safe="")
r = requests.get(f"{BASE}/similarity/{aspirin_smiles}/85.json",
                 params={"limit": 5}, timeout=30)
data = r.json()
print(f"Similar to aspirin (≥85% Tanimoto): total={data['page_meta']['total_count']}")
for m in data["molecules"][:5]:
    print(f"  {m['molecule_chembl_id']}  similarity={m.get('similarity')}")

# Substructure search
benzimidazole = quote("c1ccc2[nH]cnc2c1", safe="")
r = requests.get(f"{BASE}/substructure/{benzimidazole}.json",
                 params={"limit": 3}, timeout=30)
print(f"Benzimidazole substructure total: {r.json()['page_meta']['total_count']}")

5. Drug and Mechanism Data

import requests
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Drug record (max clinical phase, ATC class, etc.)
r = requests.get(f"{BASE}/drug/CHEMBL941.json", timeout=15)   # imatinib
drug = r.json()
print(f"Imatinib max_phase={drug.get('max_phase')}")

# Mechanisms of action — note: not every drug has mechanism records.
# Imatinib (CHEMBL941) returns 0 mechanism rows; sunitinib (CHEMBL535) has many.
r = requests.get(f"{BASE}/mechanism.json",
                 params={"molecule_chembl_id": "CHEMBL535"}, timeout=15)
for m in r.json()["mechanisms"]:
    print(f"  {m['mechanism_of_action']} → target {m.get('target_chembl_id')}")

# Therapeutic indications
r = requests.get(f"{BASE}/drug_indication.json",
                 params={"molecule_chembl_id": "CHEMBL941", "limit": 5},
                 timeout=15)
for ind in r.json()["drug_indications"]:
    print(f"  {ind.get('mesh_heading')!r}  max_phase_for_ind={ind.get('max_phase_for_ind')}")

# SVG molecular structure image — direct binary response, NOT JSON
# Do NOT call /image/{cid}.json — that endpoint raises JSONDecodeError.
r = requests.get(f"{BASE}/image/CHEMBL25.svg", timeout=15)
r.raise_for_status()
with open("aspirin.svg", "w") as f:
    f.write(r.text)
print(f"Saved aspirin.svg ({len(r.text)} bytes, looks_svg={'<svg' in r.text})")

Common Workflows

Workflow 1: Find Inhibitors for a Target

Note: pref_name__icontains matches the spelled-out name. Acronyms like 'EGFR' or 'BRAF' return 0 results — use 'epidermal growth factor receptor' or 'B-raf' (with the hyphen).

import requests, pandas as pd, time
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Step 1: Resolve the target by full name
r = requests.get(f"{BASE}/target.json",
                 params={"pref_name__icontains": "B-raf",
                         "target_type": "SINGLE PROTEIN", "limit": 5},
                 timeout=15)
targets = r.json()["targets"]
human_braf = next(t for t in targets if t["organism"] == "Homo sapiens")
target_id = human_braf["target_chembl_id"]
print(f"Using {target_id} — {human_braf['pref_name']}")

# Step 2: Paginate all potent IC50 activities (cap at 500 for demo)
url = (f"{BASE}/activity.json"
       f"?target_chembl_id={target_id}"
       f"&standard_type=IC50"
       f"&standard_value__lte=100"
       f"&standard_units=nM"
       f"&pchembl_value__isnull=False"
       f"&limit=200")
records = []
while url and len(records) < 500:
    r = requests.get(url, timeout=30)
    r.raise_for_status()
    data = r.json()
    records.extend(data["activities"])
    nxt = data["page_meta"].get("next")
    url = f"https://www.ebi.ac.uk{nxt}" if nxt else None
    time.sleep(0.2)

df = pd.DataFrame(records)
df["standard_value"] = pd.to_numeric(df["standard_value"])
print(f"Retrieved {len(df)} potent {target_id} compounds")
print(df[["molecule_chembl_id", "standard_value", "pchembl_value"]].head(10))

Workflow 2: Analyze a Known Drug

import requests
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Sunitinib (CHEMBL535) — has documented mechanisms + indications.
# Imatinib (CHEMBL941) sometimes returns 0 mechanism rows depending on ChEMBL release.
chembl_id = "CHEMBL535"

# Molecule record
m = requests.get(f"{BASE}/molecule/{chembl_id}.json", timeout=15).json()
print(f"Name: {m['pref_name']}")
print(f"MW  : {m['molecule_properties']['mw_freebase']}")

# Mechanisms
mechs = requests.get(f"{BASE}/mechanism.json",
                     params={"molecule_chembl_id": chembl_id},
                     timeout=15).json()["mechanisms"]
for mc in mechs:
    print(f"  Mechanism: {mc['mechanism_of_action']}")

# Indications
inds = requests.get(f"{BASE}/drug_indication.json",
                    params={"molecule_chembl_id": chembl_id, "limit": 5},
                    timeout=15).json()["drug_indications"]
for ind in inds:
    print(f"  Indication: {ind.get('mesh_heading')}  "
          f"(Phase {ind.get('max_phase_for_ind')})")

# Bioactivity record count
total = requests.get(f"{BASE}/activity.json",
                     params={"molecule_chembl_id": chembl_id,
                             "pchembl_value__isnull": "False",
                             "limit": 1},
                     timeout=30).json()["page_meta"]["total_count"]
print(f"Total bioactivity records (pChEMBL-tagged): {total}")

Workflow 3: SAR Study

import requests, pandas as pd, time
from urllib.parse import quote
BASE = "https://www.ebi.ac.uk/chembl/api/data"

# Step 1: Similar compounds to a lead (e.g., quinoline scaffold)
lead_smiles = "c1ccc2c(c1)cc(nc2N)c3ccc(cc3)NC(=O)c4ccccc4"
r = requests.get(f"{BASE}/similarity/{quote(lead_smiles, safe='')}/80.json",
                 params={"limit": 20}, timeout=30)
analogs = r.json()["molecules"]
print(f"Analogs found: {len(analogs)}")

# Step 2: Collect bioactivities for each analog
records = []
for compound in analogs[:20]:
    cid = compound["molecule_chembl_id"]
    acts = requests.get(f"{BASE}/activity.json",
                        params={"molecule_chembl_id": cid,
                                "standard_type": "IC50",
                                "pchembl_value__isnull": "False",
                                "limit": 20},
                        timeout=30).json()["activities"]
    for act in acts:
        records.append({
            "chembl_id": cid,
            "target": act.get("target_pref_name"),
            "IC50_nM": act.get("standard_value"),
            "pchembl": act.get("pchembl_value"),
            "mw":    (compound.get("molecule_properties") or {}).get("mw_freebase"),
            "alogp": (compound.get("molecule_properties") or {}).get("alogp"),
        })
    time.sleep(0.2)

df = pd.DataFrame(records)
if not df.empty:
    df["IC50_nM"] = pd.to_numeric(df["IC50_nM"])
    print(df.groupby("target")["IC50_nM"].describe())

Common Recipes

Recipe: Virtual Screening Filter (Lipinski rule-of-5)

import requests
BASE = "https://www.ebi.ac.uk/chembl/api/data"
r = requests.get(f"{BASE}/molecule.json",
                 params={"molecule_properties__mw_freebase__range": "300,500",
                         "molecule_properties__alogp__lte": 5,
                         "molecule_properties__hba__lte": 10,
                         "molecule_properties__hbd__lte": 5,
                         "molecule_properties__num_ro5_violations": 0,
                         "limit": 1},
                 timeout=15)
print(f"Drug-like candidates: {r.json()['page_meta']['total_count']}")

Recipe: Paginate Activities to CSV

import requests, pandas as pd, time
BASE = "https://www.ebi.ac.uk/chembl/api/data"

url = (f"{BASE}/activity.json"
       f"?target_chembl_id=CHEMBL203"
       f"&standard_type=IC50"
       f"&pchembl_value__isnull=False"
       f"&limit=500")
all_acts = []
while url:
    r = requests.get(url, timeout=60)
    r.raise_for_status()
    data = r.json()
    all_acts.extend(data["activities"])
    nxt = data["page_meta"].get("next")
    url = f"https://www.ebi.ac.uk{nxt}" if nxt else None
    time.sleep(0.3)

df = pd.DataFrame(all_acts)
df.to_csv("egfr_activities.csv", index=False)
print(f"Exported {len(df)} records → egfr_activities.csv")

Recipe: Robust Session with Retries

import requests
from requests.adapters import HTTPAdapter
from urllib3.util.retry import Retry

def chembl_session(retries=3, backoff=1.0):
    s = requests.Session()
    s.headers.update({"Accept": "application/json"})
    s.mount("https://", HTTPAdapter(max_retries=Retry(
        total=retries, backoff_factor=backoff,
        status_forcelist=[429, 500, 502, 503, 504],
        allowed_methods=["GET"])))
    return s

session = chembl_session()
r = session.get("https://www.ebi.ac.uk/chembl/api/data/molecule/CHEMBL25.json", timeout=15)
print(r.json()["pref_name"])

Recipe: Download SVG Structure Image

import requests
r = requests.get("https://www.ebi.ac.uk/chembl/api/data/image/CHEMBL25.svg", timeout=15)
r.raise_for_status()
with open("aspirin.svg", "w") as f:
    f.write(r.text)

Key Parameters

Parameter	Endpoint	Default	Description
limit	all list endpoints	20	Page size; max 1000
offset	all list endpoints	0	Pagination offset (or follow page_meta.next)
format	all endpoints	json (via .json suffix)	Also .xml, .yaml
pref_name__icontains	/target, /molecule	—	Substring on full name; acronyms don't match, use full term
target_chembl_id	/activity	—	E.g., CHEMBL203 (EGFR), CHEMBL240 (D2 receptor)
molecule_chembl_id	/activity, /mechanism, /drug_indication	—	E.g., CHEMBL25 (aspirin)
standard_type	/activity	—	IC50, Ki, Kd, EC50
standard_value__lte	/activity	—	Max activity value (paired with standard_units)
pchembl_value__isnull	/activity	—	"False" to require pChEMBL-tagged data
target_type	/target	—	SINGLE PROTEIN, PROTEIN COMPLEX, ORGANISM, …
{tanimoto} (path)	/similarity/{smiles}/{tanimoto}	—	0–100 Tanimoto threshold
{smiles} (path)	/similarity, /substructure	—	URL-encoded SMILES (urllib.parse.quote(s, safe=""))

Troubleshooting

Problem	Cause	Solution
pref_name__icontains=EGFR (or BRAF) returns 0	ChEMBL stores spelled-out names; acronyms don't match	Use "epidermal growth factor receptor"; for BRAF use "B-raf" with the hyphen
mechanism.json?molecule_chembl_id=CHEMBL941 returns empty	Not every drug has mechanism rows in every release (e.g., imatinib has 0 in current data)	Use CHEMBL535 (sunitinib) or CHEMBL192 (sildenafil) as known-populated examples
JSONDecodeError on /image/{cid}.json	The image endpoint is binary, not JSON	Always use .svg or .png suffix: /image/{cid}.svg
404 on /molecule/{id}	Invalid ChEMBL ID format	IDs must include the prefix: CHEMBL25, not 25
400 on similarity search	Unencoded SMILES (/ collides with URL path)	URL-encode: urllib.parse.quote(smiles, safe="")
Empty next page but total_count higher	Reached internal limit (typically 10000 with offset pagination)	Narrow filters (date range, target class) and re-paginate; or use the ChEMBL FTP downloads for >100K records
HTTP 429 Too Many Requests	Burst pace	Add time.sleep(0.3); mount a Retry adapter (see Recipe)
Mixed units in activity records	Different assays report in nM / µM / % inhibition	Filter standard_units="nM" and prefer pchembl_value for cross-assay comparison
data_validity_comment is non-empty	Curation flag (e.g., "Potential transcription error", "Outside typical range")	Drop these rows before SAR/regression analysis
Duplicate activity records	Same measurement reported in multiple sources	Check potential_duplicate=True and dedupe

Best Practices

• Use pchembl_value for cross-study comparisons — it normalizes IC50/Ki/EC50 to a comparable -log10 scale.
• Always check data_validity_comment before computing aggregates — flagged rows can skew distributions.
• Pin standard_units="nM" in activity queries to avoid mixing nM with µM.
• Follow page_meta.next for pagination instead of incrementing offset manually — the URL already carries the right cursor.
• URL-encode SMILES in path-style endpoints (/similarity/{smiles}/..., /substructure/{smiles}) with urllib.parse.quote(smi, safe="").
• Use a Session with retry adapter for batch work (see Recipe) — ChEMBL handles a fair amount of traffic and occasionally returns 502/503.
• For >100K records prefer the ChEMBL FTP downloads over paginated API calls.
• Be deliberate about acronyms in pref_name__icontains — EGFR, BRAF, HER2 all return 0 hits. Use the spelled-out term or filter via target_components__accession=<UniProt> instead.

Related Skills

• rdkit-cheminformatics — SMILES manipulation, fingerprints, descriptors
• datamol-cheminformatics — molecular preprocessing & featurization
• pubchem-compound-search — alternative compound database (NIH; broader coverage but less bioactivity depth)
• pdb-database — 3D structures of ChEMBL targets via RCSB PDB REST API
• opentargets-database — links ChEMBL drug-target evidence to disease associations

References

• ChEMBL website: https://www.ebi.ac.uk/chembl/
• REST API root: https://www.ebi.ac.uk/chembl/api/data/
• API docs: https://www.ebi.ac.uk/chembl/api/data/docs
• Interface docs (Django filter syntax): https://chembl.gitbook.io/chembl-interface-documentation/web-services
• Bulk downloads (for >100K records): https://chembl.gitbook.io/chembl-interface-documentation/downloads
• For SDK-based usage, see the chembl_webresource_client PyPI package; this SKILL.md uses the underlying REST API directly so no SDK install is needed.