PHI-base: Towards AI-assisted curation and FAIR pathogen-host interaction knowledge for agricultural, human and animal pathogens

The Pathogen-Host Interactions Database (PHI-base) is a manually curated, multi-species knowledge base that captures experimentally-supported pathogenicity, virulence and effector phenotypes across plant, animal, human, insect and fungal hosts. Each record links a pathogen gene or allele to a host, assay type, phenotype outcome and evidence, underpinned by PHIPO and other OBO Foundry ontologies, enabling machine-readable, FAIR data integration with genomic and multi-omics resources. PHI-base version 5.2 (https://phi5.phi-base.org/) contains 33,498 curated pathogen-host interactions involving 303 pathogenic species and 237 hosts, and 10,300 genes curated from 5,222 peer-reviewed publications. These interactions involve fungal, bacterial and protist pathogens. PHI-base is used for comparative genomics/gene function transfer, resistance target discovery and functional genome annotation in agriculture and infectious disease research. To scale our curation to meet demand, we are developing AI-assisted workflows that combine large language models (LLMs) with AI curation tools. Prototype tools support literature triage, extraction of pathogen and host entities, mapping to controlled vocabularies, and suggesting interaction types and phenotypes for biocurator review. For users, we are designing AI-enhanced interfaces for exploratory analysis, including context-aware searches of pathogen, host and phenotype terms, and interaction summaries for genes of interest. We integrate with external resources (UniProtKB, Ensembl, FungiDB) to display PHI-base knowledge within interactive genome browsers and analysis platforms. We will maintain biocurator-in-the-loop accuracy while increasing throughput, delivering knowledge to inform experiments and prioritise target genes for further analysis. Here, we will present current data content, usage patterns and initial AI workflows.