AI for Extracting Pre-Analytical Variability Data from Biomedical Literature: Feasibility and Validation

Area of Science:

• Biomedical research
• Data science
• Scientific literature analysis

Background:

• Pre-analytical variability significantly impacts biological sample research quality and reproducibility.
• Standardized reporting of pre-analytical conditions is limited, hindering systematic evaluation.
• This study investigates Large Language Models (LLMs) for structured data extraction.

Purpose of the Study:

• To explore the potential of LLMs for extracting pre-analytical variability data.
• To evaluate LLM performance in understanding and structuring complex experimental conditions.
• To assess the effectiveness of LLMs in biomedical data extraction.

Main Methods:

• Utilized a standardized parameter catalog for LLM evaluation.
• Employed specially designed prompts to test various LLMs.
• Assessed performance based on contextual understanding and structured output generation.

Main Results:

• Several LLMs (e.g., GPT-4.5, DeepSeek R1) showed strong performance in contextual understanding and CSV output.
• Consistent semantic mapping of complex conditions like storage time and temperature presented challenges.
• Targeted token reduction was found to significantly improve extraction quality.

Conclusions:

• LLMs are effective tools for supporting structured data extraction in biomedical research.
• Current LLM limitations in reproducibility and contextual fidelity necessitate expert oversight.
• Further development is needed to address challenges in mapping complex experimental parameters.