Fig. 5

A Despite significant sequence differences, the generated proteins exhibit remarkable structural consistency compared to that of the proteins in the dataset, which is a noteworthy result. It is important to highlight that the training process relies solely on topological structure labels, as no other structural information is provided. Structure ID and seq ID correspond to the maximum structural consistency and sequence consistency in the dataset, respectively. B Sequence alignment of sequences in (A). The above is the sequence from the RifDock backbone library, and the following is the sequence we designed. The identical residues are denoted in grey, while the distinct residues are highlighted in red. Concurrently, we have annotated the secondary structure type corresponding to each residue, as deduced from DSSP, where G signifies a \(3_{10}\) -helix, H represents an \(\alpha\) -helix, and T indicates a hydrogen-bonded turn