The significance of protein–protein interactions (PPIs) in biology cannot be overstated. Approximately 80% of proteins interact with other proteins to perform their essential functions [1]. Disrupting these interactions, whether through mutations in interface residues, conformational changes, or the absence of binding partners, can lead to cellular pathways malfunction and diseases, including cancer and neurodegeneration. Infections by bacteria and viruses often involve interactions with host cell-surface receptors, followed by interactions with host proteins to evade the immune response and exploit host machinery for replication and spread. Understanding these interactions and elucidating the spatial structures of protein complexes is crucial for unraveling the mechanisms behind genetic and infectious diseases and developing effective treatment strategies. Furthermore, PPI studies offer shortcuts to learning about the function of poorly characterized proteins because the function of one protein can often be deduced if its binding partner is known. Thus, both basic science and biomedical applications have attracted significant attention to investigating PPIs.

Historically, knowledge about PPIs has been derived mainly from in vitro and in vivo experiments. However, recent breakthroughs in the computational methods enable us to predict PPIs with accuracy comparable to experimental approaches, particularly for stable and even transient interactions [2∗∗, 3∗∗, 4∗, 5∗∗]. This progress is primarily attributable to the wealth of evolutionary information available in the form of millions of homologous sequences that coevolve in interacting protein pairs. Statistical and machine learning algorithms have been developed to decode these covariation signals, enabling the prediction of physiologically relevant interactions. Additionally, artificial intelligence (AI)-based models for protein structures have emerged as a game-changer, promising highly accurate 3D models of PPIs on a proteome-wide scale.

This review explores recent breakthroughs in protein structure prediction that fuel proteome-wide PPI predictions (Figure 1). We will also overview the potential biomedical applications enabled by these PPI predictions, including small molecules targeting PPIs and protein-binder design, and elucidate the imminent challenges that the community may encounter.