Large language models demonstrate surprising capability to infer causal relationships between biological entities without task-specific training.

• LLMs can identify direct causal relationships between genes, proteins, and other biomolecules
• Models achieved strong performance in a zero-shot setting across biological contexts
• Framework provides systematic evaluation of LLMs for inferring biological causality
• Potential to accelerate discovery by supplementing traditional experimental methods

This research offers a promising computational approach for mapping complex biological networks, potentially reducing time and resources needed for experimental validation in drug discovery and disease research.

Large Language Models for Zero-shot Inference of Causal Structures in Biology