Rethinking scale in AI-driven genomic medicine – The role of small biobanks

Artificial intelligence is rapidly advancing genomic medicine, but its clinical trustworthiness cannot be secured by larger datasets alone. This perspective argues that small biobanks, including national, regional, hospital-linked and disease-focused collections provide essential stress tests for AI-driven genomic medicine because they expose failures in population calibration, rare-variant interpretation, phenotype realism, privacy protection, and governance. Rather than serving primarily as substrates for training general-purpose models, small biobanks are most valuable as environments for external validation, local calibration, interpretability, federated analysis, and accountable deployment. Their local representativeness, clinical linkage, and governance structures can help determine whether AI predictions remain valid and clinically useful outside the large datasets on which they were developed. Trustworthy AI-powered genomic medicine will therefore depend not only on larger models and larger datasets, but also on smaller, well-governed biobanks that force those models to prove their validity in real-world settings.