Fig. 1: The framework of the DeepTMP.
From: Deep transfer learning for inter-chain contact predictions of transmembrane protein complexes
a The overview of the deep learning network. The input features are composed of receptor, ligand and complex features. The features of receptor are the same as ligand which includes PSSM, DCA, ESM-MSA-1b and intra-protein distance features. The features of complex exclude distance information. b The transfer learning protocol that includes the two stages of initial training and transfer learning. c The ResNet-Inception module. d The triangle-aware module that includes one triangle update, two triangle self-attention, and one transition.
