This study uses Large Language Models (LLMs), particularly GPT-4V and GPT-4 Turbo, in annotating biomedical figures, focusing on cellular senescence.
- Human annotation: 240 relations
- LLM annotation Step 1: Extract node label from figures Step 2: Determine node type(process, molecues/genes, chemical compounds, cell, disease). Step3: Determine source, target of edge Step4: Determine edge type (promote/suppress)
- Evaluation of the performance of LLM: precision, recall, and F1 score metrics for steps 1 and 3 and accuracy for steps 2 and 4.
We would like to thank the participants at BLAH8 for their collaboration and constructive advice. We are grateful to the organizers for providing this opportunity.
- GPT-4 with Vision: https://platform.openai.com/docs/guides/vision