As the participants list has now reached the full capacity of the meeting room, we can no longer accept registrations. 

The biological activity of proteins within the cellular environnement relies on their ability to insert themselves into complex interaction networks that involve several macromolecular partners, which can be others proteins, but also smaller peptides, nucleic acids or lipids. As a consequence, deciphering the protein social network remains a key issue in our understanding of the cell function, with important consequences for pharmaceutical developments. From the experimental point of view, proteomics studies provide us with a wealth of information regarding proteins networks both at the cellular and at the whole organism level [1]. Structural data regarding macromolecular assemblies is available in the Protein Data Bank (PDB), and the number of protein complexes structures released each year has been steadily increasing over the last decade. The increase in structural data generation relies in particular on the recent advances in cryo-EM techniques, which enabled resolving the structures of very large complexes. Complementary to experimental techniques, in silico methods address the identification of protein partners and the determination of the 3D structure of biomolecular assemblies [2, 3]. Integrative modeling techniques combine the best of both worlds [4, 5], as they include evolutionary information or experimental data into the process of protein complex building, especially when working on more challenging systems (such as flexible assemblies or membrane associated complexes [6]).

For decades, the most prominent tool to model protein complexes has been docking, where one attempts to predict the atomic structure of a protein complex based on the structure of its individual components. The CAPRI (for Critical Assessment of PRedicted Interactions, https://www.capri-docking.org/) initiative, which was created in 2001 has played a central role in stimulating development and progress in docking as well as scoring methods [7]. Over the years, dozens of research groups have had the opportunity to test the performance of their computational procedures against the blind prediction of more than 150 targets, i.e. macromolecular assemblies for which the experimental structure was provided to CAPRI prior to publication. Since the beginnings of CAPRI, the assessment of the quality of the predictions relies on criteria which evaluate how close the submitted models are to the reference crystallographic structure, i.e. the fraction of native contacts, the ligand and the interface root mean square deviations. In spring 2021, the publication of the AlphaFold2 software for structural prediction of proteins, shortly followed by the AlphaFold-Multimer version, which specifically deals with protein assemblies, highlighted the coming of age of machine learning in structural bioinformatics. However, these remarkably efficient tools, which are based on a single protein structure, still convey a static vision of protein-protein interactions that is now being increasingly questioned [8]. More and more data suggest that protein interfaces are dynamic objects, sometimes including disordered, flexible segments and conformational heterogeneity [9-12]. This aspect should be taken into account more often [13-16], in particular as dynamics plays a central part in defining the specificity of protein interactions [17].