Experimental assessment of AI-based interactome mapping.
Lambourne Luke, Yadav Anupama, Wang Yang, Desbuleux Alice, Kim Dae-Kyum, Laval Florent, Spirohn-Fitzgerald Kerstin, Cafarelli Tiziana, Pons Carles, Kovács István A, Jailkhani Noor, Schlabach Sadie, De Ridder David, Luck Katja, Botchkarev Vladimir V, Debnath Olivia, Bian Wenting, Shen Yun, Yang Zhipeng, Mee Miles W, Helmy Mohamed, Jacob Yves, Lemmens Irma, Rolland Thomas, McClain Gregory G, Coté Atina G, Gebbia Marinella, Kishore Nishka, Knapp Jennifer J, Mellor Joseph C, Memisoglu Gonen, Reimand Jüri, Tavernier Jan, Cusick Michael E, Zhong Quan, Aloy Patrick, Hao Tong, Charloteaux Benoit, Roth Frederick P, De Las Rivas Javier, Falter-Braun Pascal, Hill David E, Calderwood Michael A, Twizere Jean-Claude, Vidal Marc
Nature communications · 2026 · PMID 41935050 · 인용 2
Genotype-phenotype relationships are mediated through intricate networks of physical and functional interactions among macromolecules. Knowledge of the interactome is vital to understand and model genetics and cellular biology. Recent advances in accurately predicting tertiary protein structures using artificial intelligence (AI) approaches such as AlphaFold1 have revived the vision that the protein-protein interactome might be fully predictable through computational modeling of quaternary structures.
Here we present a comprehensive experimental framework to systematically assess the impact of AI-driven interactome predictions for yeast2 and human3. We find that the quality of high-confidence predictions is on par with established experimental approaches. However, in proteome-wide screening, the tested AI approaches underperform in the discovery of strictly novel protein-protein interactions (PPIs) compared to experimental reference interactome maps.
In particular, the yeast interactome map described here identifies >40-fold more novel PPIs than its AI counterpart. Strikingly, AlphaFold provides structural models for a substantial number of experimentally identified PPIs missed by the virtual screens. Our results suggest that, at this stage, the main contribution of AI predictions is to provide quaternary structure models for experimentally identified PPIs.