Fig. 4
From: Predicting protein network topology clusters from chemical structure using deep learning
Two different RNN architectures used: A (a) Seq2seq architecture with both perceiver and interpreter networks. The perceiver network is pre-trained using unlabelled data, to learn patterns and structures present in the data. (b) Finetuning of seq2seq network, where the perceiver network is connected to a fully connected layer for classification. B Pre-training and fine-tuning of the MolPMoFiT model. Weights from the pre-trained embedding matrix and three layers of LSTM are transferred and fine-tuned to perform the classification task at hand