journal menu
![[Figure 1]](dez5002fig1mag.jpg)
|
|
Figure 1
Overview of PETIMOT architecture. The model processes sequence embeddings (s) and motion vectors (  ) through 15 message-passing blocks. The GNN topology and edge features are defined from the input 3D coordinates. Edge features encode 3D geometrical properties such as the relative spatial relationships between residue pairs. Each block updates both s and representations by aggregating information from neighboring residues. SE(3) equivariance is achieved by computing the features of neighbors j in the reference frame of the central residue i. Three types of losses (LS, SS and IS) are computed, with prior normalization of the predictions for the IS and SS losses, and an additional orthogonalization of the predictions for the SS loss. |
![[Figure 1]](dez5002fig1.jpg)
 | STRUCTURAL BIOLOGY |
ISSN: 2059-7983
access
