Towards a Unified Framework for Protein Interaction Site Prediction Using Hierarchical Protein Representations

Protein interactions with various biomolecular partners pose a significant challenge in understanding cellular function and guiding drug discovery. Current computational methods are typically tailored for specific interaction classes, often lacking the ability to generalize across diverse interaction types. Researchers have now introduced PRIME-ISP, a structure-based framework for predicting interaction sites across a wide range of biomolecular interactions using a unified approach.

PRIME-ISP represents proteins at five different organizational levels - solvent-excluded surfaces, atoms, residues, secondary-structure elements, and complete proteins. By integrating information across these scales through bottom-up and top-down propagation, PRIME-ISP captures both local geometric features and global structural context.

This hierarchical representation allows the system to tackle chemically and geometrically distinct binding problems without the need for task-specific modifications. To assess the effectiveness of PRIME-ISP, the researchers evaluated its performance on protein, peptide, DNA, RNA, small-molecule, metal, and carbohydrate interaction-site prediction benchmarks.

PRIME-ISP consistently outperformed a recent unified sequence-based predictor and achieved results comparable to specialized methods, without requiring interaction-specific features. These findings demonstrate that hierarchical structural representations offer a flexible foundation for interaction-site prediction across diverse biomolecular interactions.

The source code and data for PRIME-ISP are publicly available at https://github.com/HySonLab/PRIME-ISP.