DeepIDDP is a deep learning based method for inter-domain distance prediction. In DeepIDDP, we firstly construct a non-redundant multi-domain training set from the MPDB library, and perform MSA search and DPMSA data enhancement based on the single-domain structure and full-length sequences; secondly, we extract three major features including inter-domain features, single-domain features and MSA features from the single-domain structure and MSA data. Then, a neural network with attention mechanism and deep residual blocks was designed to predict the distance distribution of inter-domain residual pairs. Finally, we integrate DeepIDDP into the previously developed SADA assembly method, which can further improve the model accuracy.
DeepIDDP On-line Sever
SADA domain assembly