view article
Figure 3
Clustering results on RBD-A and RBD-B shape reveal three distinct conformations of the main volumes according to their RBD/ACE2 arrangements. (a) The domain of inter­est of the density maps was masked through an automated protocol. Point clouds were generated using the topology representing network (TRN) sampler (Zhang et al., 2021View full citation) from the input of (i) the target density map to be masked and (ii) a reference volume provided with the mask on the domain of inter­est. Points in the target volume corresponding to the points in the reference mask region were found via optimal transport. Gaussian functions centered at the transported points were summed up to form a mask covering the domain of inter­est in the target density map. (b) Masks were created on RBD-A and RBD-B of the 11 main volumes through the automated masking algorithm. Two clear clusters were formed in the hierarchical trees constructed with the pairwise sliced-Wasserstein distance (SWD) for both RBD-A and RBD-B masks. (c) The main volumes were classified into three RBD conformations based on hierarchical clustering results, namely, down, one-up, and two-up RBD states. Their positions in the plane formed by PC1 and PC2 in the embedding space, and the density maps constructed from their embeddings, are shown. Particles were evenly distributed among the main volumes.

IUCrJ
Volume 13| Part 4| July 2026|
ISSN: 2052-2525
https://doi.org/10.1107/S2052252526004513