Search

Background: Ontologies are fundamental components of informatics infrastructure in domains such as biomedical, environmental, and food sciences, representing consensus knowledge in an accurate and computable form. However, their construction and maintenance demand substantial resources and necessitate substantial collaboration between domain experts, curators, and ontology experts. We present Dynamic Retrieval Augmented Generation of Ontologies using AI (DRAGON-AI), an ontology generation method employing Large Language Models (LLMs) and Retrieval Augmented Generation (RAG). DRAGON-AI can generate textual and logical ontology components, drawing from existing knowledge in multiple ontologies and unstructured text sources., Results: We assessed performance of DRAGON-AI on de novo term construction across ten diverse ontologies, making use of extensive manual evaluation of results. Our method has high precision for relationship generation, but has slightly lower precision than from logic-based reasoning. Our method is also able to generate definitions deemed acceptable by expert evaluators, but these scored worse than human-authored definitions. Notably, evaluators with the highest level of confidence in a domain were better able to discern flaws in AI-generated definitions. We also demonstrated the ability of DRAGON-AI to incorporate natural language instructions in the form of GitHub issues., Conclusions: These findings suggest DRAGON-AI's potential to substantially aid the manual ontology construction process. However, our results also underscore the importance of having expert curators and ontology editors drive the ontology generation process., (© 2024. The Author(s).)

Background: Phenotype ontologies are queryable classifications of phenotypes. They provide a widely-used means for annotating phenotypes in a form that is human-readable, programatically accessible and that can be used to group annotations in biologically meaningful ways. Accurate manual annotation requires clear textual definitions for terms. Accurate grouping and fruitful programatic usage require high-quality formal definitions that can be used to automate classification. The Drosophila phenotype ontology (DPO) has been used to annotate over 159,000 phenotypes in FlyBase to date, but until recently lacked textual or formal definitions., Results: We have composed textual definitions for all DPO terms and formal definitions for 77% of them. Formal definitions reference terms from a range of widely-used ontologies including the Phenotype and Trait Ontology (PATO), the Gene Ontology (GO) and the Cell Ontology (CL). We also describe a generally applicable system, devised for the DPO, for recording and reasoning about the timing of death in populations. As a result of the new formalisations, 85% of classifications in the DPO are now inferred rather than asserted, with much of this classification leveraging the structure of the GO. This work has significantly improved the accuracy and completeness of classification and made further development of the DPO more sustainable., Conclusions: The DPO provides a set of well-defined terms for annotating Drosophila phenotypes and for grouping and querying the resulting annotation sets in biologically meaningful ways. Such queries have already resulted in successful function predictions from phenotype annotation. Moreover, such formalisations make extended queries possible, including cross-species queries via the external ontologies used in formal definitions. The DPO is openly available under an open source license in both OBO and OWL formats. There is good potential for it to be used more broadly by the Drosophila community, which may ultimately result in its extension to cover a broader range of phenotypes.

Background: Journal articles and databases are two major modes of communication in the biological sciences, and thus integrating these critical resources is of urgent importance to increase the pace of discovery. Projects focused on bridging the gap between journals and databases have been on the rise over the last five years and have resulted in the development of automated tools that can recognize entities within a document and link those entities to a relevant database. Unfortunately, automated tools cannot resolve ambiguities that arise from one term being used to signify entities that are quite distinct from one another. Instead, resolving these ambiguities requires some manual oversight. Finding the right balance between the speed and portability of automation and the accuracy and flexibility of manual effort is a crucial goal to making text markup a successful venture., Results: We have established a journal article mark-up pipeline that links GENETICS journal articles and the model organism database (MOD) WormBase. This pipeline uses a lexicon built with entities from the database as a first step. The entity markup pipeline results in links from over nine classes of objects including genes, proteins, alleles, phenotypes and anatomical terms. New entities and ambiguities are discovered and resolved by a database curator through a manual quality control (QC) step, along with help from authors via a web form that is provided to them by the journal. New entities discovered through this pipeline are immediately sent to an appropriate curator at the database. Ambiguous entities that do not automatically resolve to one link are resolved by hand ensuring an accurate link. This pipeline has been extended to other databases, namely Saccharomyces Genome Database (SGD) and FlyBase, and has been implemented in marking up a paper with links to multiple databases., Conclusions: Our semi-automated pipeline hyperlinks articles published in GENETICS to model organism databases such as WormBase. Our pipeline results in interactive articles that are data rich with high accuracy. The use of a manual quality control step sets this pipeline apart from other hyperlinking tools and results in benefits to authors, journals, readers and databases.