Fig. 1From: Protein–protein interaction site prediction by model ensembling with hybrid feature and self-attentionThe architecture of StackingPPINet. StackingPPINet contains multiple base classifiers, named PPINets. Each PPINet consists of a feature forming module (FFMod), a feature aggregation network (FANet) and a prediction module (PPIPred). Multiple PPINets are trained independently and ensembled via StackingBack to article page