Behind the cover image of the August 18 issue of Biophysical Journal is DynMoCo, an artificial intelligence–driven framework utilizing deep learning for extracting meaningful patterns from high-dimensional molecular dynamics (MD) trajectories. By modeling each simulation frame as a molecular graph, in which residues act as nodes and physical contacts as edges, DynMoCo detects from massive atomic motions co-moving structural communities of protein conformational changes, such as the force-induced extension of the integrin family of receptors.
The image depicts the 32 dynamic communities that DynMoCo identified for αIIbβ3 under steered MD (force-ramp) simulation, rendered in VMD as color-labeled patches across the protein's surface that reveal which residues move together as the integrin extends. This force-induced reorganization is what allows a platelet to physically grip its environment, a crucial mechanotransduction step that drives both healthy clotting and pathological thrombosis. Notably, these community boundaries closely track known structural domains, indicating that each domain moves as a nearly independent rigid unit. This contrasts with the related integrin αVβ3, in which communities span multiple domains, reflecting tighter cross-domain coupling that effectively stiffens the protein and slows down the unbending process.
The research shown in this article indicates that integrin activation, often described as a switch between two end states (bent-closed and extended-open), involves an intermediate transition in which much of the interesting mechanics occurs: which structural regions move as a unit, which act as hinges, and how these groupings shift as force accumulates. By reducing a high-dimensional MD trajectory into a smaller set of interpretable, temporally tracked communities, DynMoCo allows us to address these questions directly. Because the framework is not integrin specific, it can also be applied to other mechanosensitive proteins.
To find out more about this work by Dr. Jing Li’s and Dr. Cheng Zhu’s research teams, check out this GitHub repository.
— Lingchao Mao, Mingu Kwak, Amir Hossein Kazemipour Ashkezari, Zhenhai Li, Peiwen Cong, Yunfeng Chen, Jung Hun Phee, Sooyeon Kang, Jing Li, and Cheng Zhu