Fig. 2: The model architecture of ProRefiner. | Nature Communications

Fig. 2: The model architecture of ProRefiner.

From: ProRefiner: an entropy-based refining strategy for inverse protein folding with global graph attention

Fig. 2

A partial sequence, either given or constructed from a base model’s generation, and the backbone structure are encoded to obtain the graph features. Several memory-efficient global graph attention layers are employed to propagate the graph features and learn global residue interactions. Finally the whole sequence is generated in one shot.

Back to article page