Research On Protein Domain Boundary Prediction Based On Deep Learning

Protein domain boundary division often seen as first step towards more advanced three-dimensional protein structure and function prediction work. Because the methods based on the X-ray crystallography and NMR are too costly and low efficiency, far unable to catch up with the speed that new sequence was found, the development of high accuracy prediction tool is the key to bridge this gap.Through the analysis of the existing domain prediction method and select the greatest potential method so far. We carried out the protein domain boundary prediction working based on deep learning. Mainly with its powerful features learning ability and automatic mining the helpful features that are contribute to predicting boundary. The first proposed model is the use of convolutional neural network, it modify the convolutional neural network convolution kernel and the internal structure of the network layer, making it more suitable for protein domain boundary prediction. The second proposed model combine convolution neural network and the long short term memory network to a single network,so that the long short term memory network learning the new features from output that extracting by convolutional neural network instead of original amino acid sequence.The results showed that, the precision and recall of out models are better than the models based on the traditional machine learning methods PPRODO, and the recall is much more higher than template-based method Pfam while precision is as good as its.