Fig. 2From: Protein–protein interaction site prediction by model ensembling with hybrid feature and self-attentionThe architecture of PPINet. The feature forming module (FFMod) extracts low level features from the input protein sequence. As an example, K-PseAA and PhyChem are applied in this showcase to extract the targeted residue feature and the context feature respectively. These features are then processed by a feature aggregation network (FANet). Based on the aggregated features, the predictor (PPIPred) performs classificationBack to article page