AlphaFold-Based Protein-Protein Interaction Prediction Methods Classification

Protein-protein interactions (PPIs) are fundamental to biology. While experimental PPI detection is crucial, it is often expensive, error-prone, and struggles with transient interactions. Computational, especially structure-based, methods offer valuable alternatives, their reliance on experimentally determined structures has historically restricted their applicability. AlphaFold has transformed PPI prediction by providing accurate protein structure predictions, even without experimental data, thereby overcoming a major bottleneck in PPI research. This review examines the transformative impact of AlphaFold-based methods - including AlphaFold2, AlphaFold-Multimer, and AlphaFold3 - on PPI prediction, categorizing their contributions into six methodological domains. These methods utilize AlphaFold's capabilities to directly predict interacting protein complex structures, employing enhanced metrics for accuracy and sensitivity in PPI detection based on predicted interface features .The combination of AlphaFold-generated structures with computational techniques such as machine learning and network analysis is also considered to further improve prediction reliability and coverage. Furthermore, the critical role of the AlphaFold Database (AFDB) is highlighted in facilitating large-scale PPI studies by providing access to millions of predicted structures, enabling researchers to explore interaction networks and binding interfaces at an unprecedented scale. The review also surveys databases that provide critical information on protein interactions, sequences, structures, as PPI research hinges on the effective utilization and integration of these databases.