AI‐based data synthesis of crop trait prioritization studies

Abstract Synthesis of data from crop trait prioritization studies (CTPS) can provide insights to support decision‐making, such as institutional funding allocation, and trait prioritization in crop improvement programs. This type of data synthesis is constrained by the lack of standardized crop trait terminology and suitable methods to deal with data heterogeneity. Crop trait ontologies provide terminology standardization, but annotating documents to link terms to ontology terms is time‐consuming and may therefore miss trait terminology emerging from CTPS due to a data annotation bottleneck that constrains data synthesis. Natural language processing (NLP) techniques based on large language models (LLMs) can help in extracting information from unstructured text with no manual text annotation involved. This study applied NLP to synthesize unstructured text data extracted from CTPS by a recently published scoping review. Results show that (1) the trait vocabulary diversity used in CTPS varies per crop and by gender intentionality of CTPS, (2) crop trait preferences increasingly focus on food quality and climate adaptation traits, and (3) existing crop ontologies provide a good coverage of terms found in CTPS but might require the addition of terms, especially in crops such as cassava and sweet potato. This study demonstrates the utility of applying NLP and LLM to synthesize trait preference data across crops and timescales, potentially modeling an approach for broader utility to breeding programs and crop ontology curators alike.