AI-based structural modelling and pathogenic variant mapping across the human protein interactome

AI-based modeling produced 109,282 predicted human protein interaction structures, enabling systematic interface characterization and pathogenic variant mapping across the human interactome.

• Why it matters: Most protein–protein interaction structures remain unresolved, limiting large-scale understanding of how interface changes contribute to disease mechanisms and guiding experimental follow-up.
• What they did: The team generated AI-predicted structures for 109,282 human protein interactions, then quantified interfaces using residue involvement, buried surface area, intermolecular bonding, and energetic properties.
• The result: Pathogenic variants were mapped onto these models, revealing interface-localized mutations for over 4,000 annotated disease categories and enabling disease-context analyses for acute myeloid leukemia, Alzheimer’s disease, and cardiomyopathy.

AI summary of the abstract · may contain errors · not medical advice · read the paper.