Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning?

2026-06-13 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Life Sciences & Biology · Depth: Expert, quick

Summary

A study investigates few-shot biomedical relation extraction (BioRE) using prompt-based learning with large language models (LLMs) as an alternative to supervised methods. The research compares two task formulations: pairwise classification, which predicts relations for individual entity pairs, and joint generation, which extracts multiple relations in a single model call. Experiments on the BioREDirect dataset show a precision-recall trade-off, with pairwise classification achieving higher recall and joint generation being more precise and computationally efficient. The best-performing model achieved a micro-F1 score of 0.44, significantly surpassing previous few-shot results of 0.34 but remaining below the supervised baseline of 0.56. This gap is largely due to one ambiguously defined relation type. When evaluated with macro-F1, which accounts for imbalanced settings, prompt-based approaches outperformed the supervised baseline (0.45 vs. 0.38), especially for rare relation types. These findings suggest LLMs hold potential for BioRE in low-resource environments and emphasize clear relation schemas.

Key takeaway

For NLP Engineers developing biomedical knowledge extraction systems, you should consider prompt-based LLMs for few-shot relation extraction, especially when annotated data is scarce or relation types are imbalanced. While micro-F1 scores may lag supervised baselines (0.44 vs. 0.56), LLMs excel with rare relation types and achieve a higher macro-F1 (0.45 vs. 0.38). Prioritize defining unambiguous relation schemas to maximize performance and computational efficiency.

Key insights

LLMs offer a viable few-shot alternative for biomedical relation extraction, outperforming prior few-shot methods, especially for rare relations.

Principles

• Clear relation schemas are crucial for BioRE performance.
• Macro-F1 better assesses performance in imbalanced BioRE settings.
• Joint generation is more precise and efficient than pairwise classification.

Method

The study compares prompt-based few-shot BioRE using LLMs, evaluating pairwise classification (individual entity pairs) against joint generation (multiple relations in one call) on the BioREDirect dataset.

In practice

• Consider LLMs for BioRE in low-resource scenarios.
• Prioritize well-defined relation schemas for BioRE tasks.
• Use joint generation for higher precision and efficiency.

Topics

• Biomedical Relation Extraction
• Large Language Models
• Few-Shot Learning
• Prompt Engineering
• Knowledge Graph Construction
• BioREDirect Dataset

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Measuring the sensitivity of LLM-based structured extraction to prompt, model, and schema choices in clinical discharge summaries — LLM extraction sensitivity varies significantly with schema design and model choice, especially on absence-versus-silence distinctions.
• Implementing a custom trainable component for relation extraction — Custom relation extraction can be built using spaCy, Thinc, and Hugging Face transformers for specialized NLP pipelines.
• SciDef: Automating Definition Extraction from Academic Literature with Large Language Models — SciDef automates definition extraction from scientific papers using LLMs, achieving 86.4% accuracy with optimized prompting.
• Improving Answer Extraction in Context-based Question Answering Systems Using LLMs — Fine-tuning pre-trained LLMs on benchmark QA datasets significantly improves answer extraction accuracy and contextual understanding in QA systems.
• A Larger Annotated Corpus of Portuguese Coreference — LLMs can annotate Portuguese coreference, but a new LLM-based system did not outperform existing rule-based methods.
• Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification — Prompt sensitivity and demonstration quality critically impact LLM performance in metalinguistic classification tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.