PMC Caption Pathway Extraction Audit

Methodology

Image filename
PMC1234__F2.jpg → Fetch PMC XML
efetch API → Extract Figure Caption → Regex Extraction
X signaling pathway → Match vs Gemini
word overlap + fuzzy

Source: 1,000 random samples from 49k pathway images
Extraction patterns: "X signaling", "X pathway", "X activation", known pathway names (MAPK, PI3K, etc.)
Matching: Word-based overlap with fuzzy stemming (apoptosis~apoptotic, mTOR~mTORC1)
Limitations: 28% of articles restricted by publishers; biological synonyms (SHH≠Hedgehog) not handled

Note: Captions typically describe the biology shown, not explicitly label the pathway. A match means the caption mentions the same pathway Gemini identified—not strict ground truth confirmation.