Reconstruction Of Protein Structure Based On Auto Encoder

Proteins are biological macromolecules and play an important role in the life organisms. Because the structure of the protein determines its function, accurate prediction of protein structure are greatly signi?cant for in the bio-pharmaceutical and disease research.A method of protein structure reconstruction based on Auto Encoder is designed.It uses the homologous proteins which are similar to the target protein as the training data, and reconstructs the target protein decoys. Experiments show that, for the training set of proteins our methods are capable of good reconstruction, and could be used for sampling the decoys of the target protein.This paper also discusses how the degree of homologous protein similarity and the size of training set data relate to the target protein structure reconstruction. Experiments show that a higher similarity of target protein with training set of proteins,better the reconstruction. Further experiments show that huge size of training set data is not necessary.The reconstruction model is also applied to protein-peptide docking,and the primary experiments show that it helps improving the e?ciency of proteinpeptide docking.