Skip to main content
. 2023 Oct 28;39(11):btad663. doi: 10.1093/bioinformatics/btad663

Figure 1.

Figure 1.

TT3D model architecture. TT3D follows the structure and training procedures of Topsy-Turvy, but with an augmented protein embedding. We concatenate a one-hot encoding of the Foldseek 3Di (van Kempen et al. 2023) string to the protein language model (PLM)-based embedding before passing this representation into the convolutional portion of the architecture.