A novel method for protein structure retrieval using tableau representation and sparse coding

Protein retrieval is a difficult task and has become a hot issue recently due to the complex structure and large data size of proteins. This work helps biologists investigate the link between structure and function of a protein in a deeper level and can be used in lots of biomedical applications. The retrieval system gives scores to all proteins in the database, e.g. SCOP or PDB, by given a query protein to compare with them. In this paper, we propose a novel algorithm based on sparse coding to retrieve proteins in the database using tableau representation. Both unsupervised and supervised methods are studied in the proposed algorithm where the sparse coefficient is regarded as similarity measurement. Experiments are conducted on ASTRAL 1.73 95% database and show that the proposed algorithms can improve the original feature extraction method which only uses cosine similarity.