GENET: AI-Powered Interactive Visualization Workflows to Explore Biomedical Entity Networks

Formulating hypotheses about gene-disease associations requires logical inference from prior data, followed by a laborious literature review. AI models trained on curated datasets (e.g., GWAS Catalog) can suggest SNP–disease links, but validating these predictions still demands manual evidence extraction. To streamline this process, we present GENET (Genomic Evidence Network Exploration Tool), an AI-enhanced, end-to-end visual analytics workflow applied to Age-Related Macular Degeneration (AMD). GENET comprises four sequential steps: (1) biomedical network analysis: a dual-encoder neural model identifies genes or SNPs associated with a target disease and vice versa; (2) literature evidence mining pipeline: a pipeline retrieves relevant papers and, using large language models, extracts biomedical entities and relations; (3) clustering: embeddings from pre-trained biomedical language models (BioBERT, BioLinkBERT) are clustered to group related concepts; (4) interactive visualizations: clusters and their networks are visualized with interactive features for hypothesis testing and insight generation. The workflow enables iterative hypothesis formulation and evidence validation, uncovering novel associations. GENET is open-sourced (https://github.com/BiomedSciAI/genet) and available for demonstration at https://genet.pythonanywhere.com.