OpenPharma/pharma-kb-obesity

Pharma KB — Obesity Drug Landscape

A structured pharmaceutical knowledge base covering the obesity drug pipeline — compiled from public regulatory, clinical, and scientific sources. Designed for RAG pipelines, competitive intelligence workflows, and pharma-domain LLM fine-tuning.

Overview

Phase distribution (285 drugs):

Top pipeline positions (activity score, 0–100):

• —Eli Lilly and Company — 86.84
• —Novo Nordisk — 86.44
• —Hanmi Pharmaceutical — 34.39

Sample Article

Drug articles are markdown files with structured frontmatter and narrative body:

---
title: semaglutide (subcutaneous, diabetes/obesity/NASH), Novo Nordisk
type: drug
slug: semaglutide
phase: Launched
originator: Novo Nordisk A/S
mechanism: Glucagon-like peptide 1 receptor agonist
indication_count: 10
fda_approval_count: 4
ct_trial_count: 215
targets:
- glucagon-like-peptide-1-receptor
indications:
- obesity
- non-insulin-dependent-diabetes
- metabolic-dysfunction-associated-steatohepatitis
# ... more fields
---

The body covers mechanism of action, clinical trial history, regulatory milestones (FDA/EMA/PMDA), competitive context, deal activity, and preprint evidence.

Dataset Structure

pharma-kb-obesity/
├── indications/
│   └── obesity.md              # Landscape overview, pipeline phases, deal velocity
├── drugs/                      # 277 drug articles — all phases
│   ├── semaglutide.md
│   ├── tirzepatide.md
│   ├── cagrilintide.md
│   └── ...
├── companies/                  # 248 company articles with pipeline positions
│   ├── eli-lilly.md
│   ├── novo-nordisk.md
│   └── ...
├── targets/                    # 35 target articles
│   ├── glucagon-like-peptide-1-receptor.md
│   ├── gastric-inhibitory-polypeptide-receptor.md
│   └── ...
├── MANIFEST.md                 # Resolve URLs for all 569 articles (for direct fetch)
└── data/
    └── train.parquet           # Structured index — one row per entity

Each article is a markdown file with YAML frontmatter and a narrative body covering mechanism, clinical status, regulatory history, and competitive context.

Usage

Load the index

from datasets import load_dataset

ds = load_dataset("OpenPharma/pharma-kb-obesity", split="train")
df = ds.to_pandas()

# Filter: Phase 3 drugs
phase3 = df[(df["entity_type"] == "drug") & (df["phase"] == "Phase 3")]
print(phase3[["slug", "title", "mechanism", "originator"]])

Fetch any article directly

Every article has a stable resolve URL:

https://huggingface.co/datasets/OpenPharma/pharma-kb-obesity/resolve/main/companies/eli-lilly.md
https://huggingface.co/datasets/OpenPharma/pharma-kb-obesity/resolve/main/drugs/tirzepatide.md
https://huggingface.co/datasets/OpenPharma/pharma-kb-obesity/resolve/main/indications/obesity.md

A full listing of all 569 resolve URLs is available at `MANIFEST.md`.

Clone for full article access

git clone https://huggingface.co/datasets/OpenPharma/pharma-kb-obesity

RAG pipeline

from huggingface_hub import snapshot_download
from pathlib import Path

repo_path = Path(snapshot_download("OpenPharma/pharma-kb-obesity"))

# Load all articles as documents
documents = []
for md_file in repo_path.rglob("*.md"):
    documents.append({
        "text": md_file.read_text(),
        "source": str(md_file.relative_to(repo_path)),
    })

# Feed into LlamaIndex, LangChain, or any embedding pipeline

Read a specific article

import yaml
from pathlib import Path

def parse_article(path):
    content = Path(path).read_text()
    _, fm_raw, body = content.split("---", 2)
    return yaml.safe_load(fm_raw), body

fm, body = parse_article(repo_path / "drugs/semaglutide.md")
print(fm["phase"], fm["mechanism"])
print(body[:500])

What's in Each Article

Drug articles include:

• —Phase, mechanism of action, originator, indication coverage
• —FDA approval history (approvals, label cross-checks)
• —Clinical trial counts (ClinicalTrials.gov cross-referenced)
• —Regulatory conflicts and data quality flags
• —Aliases (brand names, INN variants, trial codes)
• —Industry research attention — rolling 7-day attention rank and relative research activity vs. peers across the obesity landscape

Company articles include:

• —Drug count and phase distribution per indication
• —Activity score and tier across indications
• —Deal counts and partnership history
• —Investment Profile — public/private status, ticker (exchange + symbol), total capital raised, latest funding round, key investors, and recent pipeline/deal developments sourced from public financial data
• —Competitive drug watchlist — the drugs this organization viewed most over the trailing 7-day window (predominantly other companies' drugs, surfacing competitive interest)

Target articles include:

• —Gene symbol, protein family, disease associations
• —Drugs targeting this mechanism with phase summary
• —Literature and patent coverage notes

Indication article includes:

• —Full pipeline phase distribution and trend analysis
• —Company activity rankings with scores
• —Mechanism class distribution (GLP-1 RA, GIP RA, GCGR, amylin, etc.)
• —Deal velocity and top deal activity by type
• —Industry research activity — ranked heatmap of research attention across top drugs in the landscape for the trailing 7-day window

Structured Index Schema

data/train.parquet — one row per entity (569 rows total):

Data Sources

This dataset is synthesized from the following public sources:

• —ClinicalTrials.gov — trial counts, recruitment status, phase mapping
• —FDA — drug approvals, labels, FAERS, patents
• —EMA — European approvals and regulatory milestones
• —PubMed — scientific literature
• —ChEMBL — mechanism of action, ADMET, structure data
• —bioRxiv / medRxiv — preprints
• —UniProt — target biology
• —Open Targets — disease-target associations

All data is synthesized and structured — not a direct export from any single source.

Releases

More Indications

Additional indication datasets (MASLD/NASH, cardiovascular, oncology) are in progress under the OpenPharma organization.

License

CC BY-NC 4.0 — free for research and non-commercial use with attribution.

Citation

If you use this dataset in your work, please cite:

@dataset{pharma-kb-obesity-2026,
  title   = {Pharma KB — Obesity Drug Landscape},
  year    = {2026},
  url     = {https://huggingface.co/datasets/OpenPharma/pharma-kb-obesity},
  license = {CC BY-NC 4.0}
}