A novel pipeline for the rapid expansion of ecological trait databases using LLMs

Abstract

This paper presents a novel workflow leveraging Large Language Models (LLMs) to rapidly extract trait data from fungal species descriptions, addressing a significant bottleneck in ecological research. We developed and evaluated an LLM pipeline to extract morphological trait data from arbuscular mycorrhizal fungi, comparing performance against a manually curated dataset (TraitAM). Results demonstrate the potential of LLMs for automated trait data acquisition, though accuracy varies by trait and model, with systematic biases observed. This framework offers a blueprint for building trait databases across diverse taxa and domains, significantly accelerating ecological research and conservation efforts.