megaMine

Hybrid literature mining for precision oncology.

megaMine extracts gene-drug-cancer therapeutic evidence from PubMed, normalizes cancer labels, refines resistance evidence, tracks temporal trends, detects contradictions, links to ClinicalTrials.gov, and exports a provenance-aware knowledge graph with an interactive HTML report.

Literature-derived evidence only. Not a clinical treatment recommendation.

---

What it does

Given a PubMed query, megaMine produces:

• Normalized gene-drug-cancer evidence rows
• 3-tier resistance refinement (observed → context → direct evidence)
• Temporal trend classification per gene-drug-cancer triplet
• Contradiction detection with conflict scores
• ClinicalTrials.gov linkage
• Knowledge graph (GraphML + CSV)
• Standalone interactive HTML report (no internet required)

---

Installation

conda create -n megamine python=3.9 -y
conda activate megamine
pip install "git+https://github.com/Junaid13913/megaMine.git"

---

Quick start

megaMine \
  --q "EGFR AND erlotinib AND resistance AND NSCLC" \
  --years 2020-2024 \
  --max-records 500 \
  --email "your@email.com" \
  --ncbi-api-key "YOUR_KEY" \
  --require-gene-and-drug \
  --require-known-drug \
  --year-binned \
  --out my_run

Output files:

File	Contents
my_run.xlsx	Evidence rows, temporal trends, contradictions, trials
my_run_graph_nodes.csv	Knowledge graph nodes
my_run_graph_edges.csv	Knowledge graph edges
my_run_graph.graphml	Graph for Cytoscape / Neo4j
my_run_HTML_REPORT.html	Interactive report — open in any browser

---

Requirements

• Python ≥ 3.9
• NCBI email + API key (free at https://www.ncbi.nlm.nih.gov/account/)
• Internet access for PubMed, Europe PMC, PubTator, HGNC

---

Case study

Query: EGFR AND erlotinib AND resistance AND NSCLC · 1,000 papers · 2015–2024

• 1,150 extracted rows → 210 verified evidence rows from 166 unique PMIDs
• 13 canonical cancer types after normalization
• 14 unique drugs detected
• 45 temporal triplets — 1 rising resistance signal (EGFR + erlotinib + NSCLC)
• 436 graph nodes · 1,587 edges
• 23 ClinicalTrials pairs

---

Getting help

megaMine --help
megaMine --list-cancers

---

Author

Muhammad Junaid; Ajou Precision Medicine Lab (APML), Ajou University, School of Medicine, South Korea; junaidm@ajou.ac.kr