Fig. 2From: Multi-head attention-based U-Nets for predicting protein domain boundaries using 1D sequence features and 2D distance mapsThe overall architecture of DistDom. The 2D distance maps and 1D features are processed by a 2D U-Net and 1D U-Net respectively. The 2D U-Net output is converted by the 1D adaptive max pooling to the output of \(L \times 1\). The \(L \times 1\) outputs from the 1D U-Net and 2D-U-Net are combined together as the hidden features for the multi-head attention layer to predict the probability of being in a domain boundary or notBack to article page