USDA Phytochemical Database — Enriched v2.3 (400-Row Sample)

A 400-record sample of the USDA Dr. Duke's Phytochemical and Ethnobotanical Database, denormalized into a flat 10-column schema and enriched with quantitative signals from five sources:

pubmed_mentions_2026: PubMed publication count per compound (NCBI E-utilities)

clinical_trials_count_2026: ClinicalTrials.gov v2 study count per compound

chembl_bioactivity_count: ChEMBL v35 bioassay data points (CC BY-SA 3.0)

patent_count_since_2020: USPTO patents since 2020-01-01 (PatentsView REST API)

pubchem_cid + canonical_smiles: PubChem compound identifier and canonical SMILES notation (PubChem REST API)

Schema: chemical, plant_species, application, dosage, pubmed_mentions_2026, clinical_trials_count_2026, chembl_bioactivity_count, patent_count_since_2020, pubchem_cid, canonical_smiles

Records: 400 (top compounds by PubMed mentions). Total dataset: 76,907 records across 24,746 compounds and 2,313 species. Full dataset: https://ethno-api.com

Formats: JSON (~25 MB) + Parquet (~800 KB, Snappy compression). Methodology: https://github.com/wirthal1990-tech/USDA-Phytochemical-Database-JSON/blob/main/METHODOLOGY.md

Changes in v2.3 vs v2.2:

Deployed a custom nomenclature canonicalization engine to resolve legacy 1990s USDA formatting (e.g., hyphenated strings) that previously caused a ~28% drop rate in standard API lookups.

Rescued 997 "lost" SMILES and PubChem CIDs, significantly improving structural coverage for machine learning pipelines.

Introduced the Patent-Literature Gap mapping: Identified 994 compounds exhibiting high patent activity but low academic publication counts.