Inside the cell, proteins rarely act alone. They assemble into dynamic complexes that perform essential functions, from gene expression to stress responses. Yet, precisely deciphering these assemblies remains a major challenge: experimental approaches are time-consuming, and the number of possible protein combinations quickly becomes overwhelming.
A study published in Acta Crystallographica D and involving Kamel Hammani (IBMP) proposes an original solution based on recent advances in artificial intelligence. The approach, named Alphafuser, combines AlphaFold-based structural predictions with experimental interaction data to progressively reconstruct higher-order protein complexes.
The idea is simple yet powerful: rather than testing all possible combinations, a problem that rapidly becomes intractable, Alphafuser iteratively discards unlikely associations based on a confidence score of protein–protein interfaces. This strategy focuses computational efforts on the most relevant complexes, making large-scale exploration of interactomes finally feasible.
Beyond prediction, the authors experimentally validated several interactions identified by the method, confirming Alphafuser’s ability to reveal biologically meaningful assemblies. This approach opens new perspectives for understanding the functional organization of cells by directly linking interaction networks to molecular structures.
By combining artificial intelligence with experimental biology, this work perfectly illustrates a major shift in the field: using computation not to replace experiments, but to better guide them.
