Dataset Card for Aggregated ChEMBL Target-SAR Matrix (lovingscience/chembl-v37-target-sar)

This dataset is a vectorized, aggregated Structure-Activity Relationship (SAR) matrix derived from ChEMBL Release 37, linking target-ligand pairs with primary bioactivity values and classification bounds.

Data Provenance & Base Filters

This dataset is built on top of chembl-raw-activities, meaning all underlying measurements have already passed the following baseline filters during extraction:

• Assay Confidence Score: confidence_score >= 7 (Direct target-assay mapping)
• Data Validity: Excluded erroneous comments (Outside typical range, Non-standard unit, Author error, Unlikely value)
• Duplicates: Excluded potential_duplicate == 1
• Missing Data: Excluded rows lacking a valid pChemBL value (or those unable to be recalculated from standard values).

Aggregation & Endpoint Hierarchy Rules

• ChEMBL Release Version: 37
• Primary Endpoint Selection Hierarchy: Kd > Ki > IC50 > EC50 > AC50 (highest confidence endpoint selected per pair)
• Activity Classification:
  • active_measured: Exact = measurement available.
  • inactive: Flagged inactive for structural relation > 10 µM (pChemBL <= 5.0).
  • active_censored: Flagged highly potent for structural relation < 1 nM (pChemBL >= 8.0).
• Total Unique Target-Ligand Pairs: 3,022,174

Activity Classification Breakdown

Activity Class	Count
active_measured	2,618,182
inactive	379,604
active_censored	24,388

Selected Primary Endpoint Breakdown

Primary Endpoint	Count
IC50	1,254,639
Potency	883,939
Ki	456,365
EC50	146,413
AC50	145,722
Kd	134,648
ED50	411
XC50	37

Single Protein Targets

• Pairs With At Least One Confidence-9 Measurement: 2,217,209 (73.4%)
• For target-level "is this a single protein" filtering, use target_type == 'SINGLE PROTEIN', not is_single_protein - the latter reflects assay curation confidence for this specific pair's measurements, not the target's own classification.

Column Schema

• target_chembl_id (string): Target ChEMBL ID.
• inchikey (string): Standardized InChIKey (27 characters) of largest organic fragment, matching chembl-smiles-curated.
• standard_inchikey (string): Raw ChEMBL standard InChIKey.
• canonical_smiles (string): Isomeric canonical SMILES string.
• molecule_chembl_id (string): Molecule ChEMBL ID.
• primary_endpoint (string): Selected primary endpoint type (Kd, Ki, IC50, EC50, AC50).
• activity_class (string): Classification (active_measured, inactive, active_censored).
• pchembl_mean (float32): Mean pChemBL value over all exact measurements for this pair.
• pchembl_median (float32): Median pChemBL value for primary endpoint.
• pchembl_std (float32): Standard deviation of pChemBL measurements.
• pchembl_min, pchembl_max (float32): Min and Max measured pChemBL values.
• pchembl_lower_bound, pchembl_upper_bound (float32): Censoring bounds for inequality measurements.
• n_measurements (uint32): Count of exact measurements aggregated into this pair.
• target_type (string): Target-level classification (e.g. SINGLE PROTEIN) - the correct field for single-protein filtering.
• is_single_protein (bool): True if any exact measurement for this pair came from a confidence-9 ("Direct single protein target assigned") assay. Per-pair assay-confidence signal, not a target classification.
• target_organism, uniprot_accession, target_pref_name: Target annotations.
• mw_freebase, alogp, hba, hbd, psa, qed: Computed molecular descriptors.