Human Protein-Protein Interaction (PPI) prediction server

Protein-Protein interaction(PPI) plays a critical role in cellular biological process, it is thus important to identify PPI with high-throughput and high quality to understand the function of proteins, disease occurrence, and design therapy.

Our Human PPI prediction server is a sequence based human protein-protein interaction prediction server. The model is based on Stacked auto encoder(SAE), one of the deep learning algorithms. Autoencoder is a kind of artificial neural network that applies unsupervised learning algorithm. Unsupervised learning algorithm infers a function to construct hidden structure from unlabeled data, in other words, it tries to make the output similar to input , which is a encoding-decoding process. A stacked autoencoder consists of multiple layers of autoencoders, which are layerwisely trained in turn, and the output of the former layer is wired to the inputs of the successive layer.

To use the Human PPI prediction server, all that the users need to provide are the human protein pair sequences. Once the sequences are submitted, they are encoded by Autocorrelation(AC) method and predicted by our trained model. The users will get the results with '✔' standing for €˜interaction while '✘' standing for predicted €˜no interaction.

Some of the dataset mentioned in the paper could be found at https://github.com/pkumdl/ppi_dataset

Reference: Sun T, Zhou B, Lai L, et al. Sequence-based prediction of protein protein interaction using a deep-learning algorithm[J]. BMC bioinformatics, 2017, 18(1): 277.