Abstract
Abstract
Catalytic metal-binding sites are established drug targets for many human diseases, yet systematic study of how metal function relates to disease is limited by the lack of a resource combining structural localization of human metal-binding sites with catalytic annotation at scale. We introduce HEAM (the Human Enzymatic Annotation of Metalloproteins), a structurally annotated dataset of 1,933 human metalloenzymes and 1,171 non-catalytic metalloproteins. HEAM incorporates annotated metals into protein structures at their corresponding binding sites and classifies each site as catalytic or non-catalytic using the machine learning classifier MAHOMES II. HEAM more than quadruples the human site-level coverage of the largest prior metalloenzyme database.
Using HEAM, we show that human metalloproteins are significantly enriched in KEGG disease pathways relative to the human proteome, and that metalloenzymes specifically are significantly enriched in ClinVar pathogenic variants — a distinction obscured in prior, catalytically unresolved metalloprotein datasets. These pathogenic variants are concentrated in metabolic disorders, indicating that metabolic dysfunction is a substantial contributor to metalloenzyme-linked disease. By pairing 3D structural annotation with catalytic classification, HEAM provides both a foundation for structure-guided drug design against metal-binding sites and a resource for determining how catalytic and non-catalytic metal function contribute to human disease.