Protein structure prediction using bidirectional neural networks
Chen, Jin Miao
Date of Issue2007
School of Computer Engineering
This thesis is focused on protein secondary structure (PSS) prediction which is one of the most important problems in bioinformatics. Most of existing prediction methods either use a sliding fixed-sized window centered on the residue of interest or train on single positions of the sequence to classify mutually independent secondary structures. They are unable to take into account long-range interactions between amino acids and strong correlations between secondary structure (SS) elements. In this thesis, we propose and develop three novel prediction models, namely bidirectional long short-term memory (BLSTM), bidirectional segmented-memory recurrent neural network (BSMRNN) and cascaded bidirectional recurrent neural network (Cascaded BRNN), which are all variants of bidirectional recurrent neural network (BRNN).
DRNTU::Engineering::Computer science and engineering::Computer applications::Life and medical sciences