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

The human proteome consists of proteins that interact physically to coordinate various cellular functions. However, the three-dimensional structures of most protein interactions are still unknown. Researchers conducted an artificial intelligence-based modeling of 109,282 human protein interactions, creating a structural resource for further analysis.

This modeling generated predictions of protein interaction structures based on various factors, including interface residue involvement, buried surface area, intermolecular bonding, and energetic properties.

The team also mapped data on pathogenic variants, or mutations associated with disease, onto these predicted structures. This allowed them to identify mutations that occur in protein interaction sites for over 4,000 annotated disease categories. They examined representative diseases such as acute myeloid leukemia, Alzheimer's disease, and cardiomyopathy to explore the distribution of these interface mutations within protein interaction networks.

The researchers also assessed the strengths and limitations of their predictive capabilities, including comparisons with experimental protein structures. The resulting structural resource, along with the analyses conducted, provides a foundation for generating new hypotheses and guiding experimental studies involving human protein interactions in both normal physiological processes and disease contexts.