Protein Structure Prediction by Memory-Base Reasoning

摘要

Memory-based reasoning (MBR) is a technique that makes intensive use of memory to recall some specific episodes from the past for problem solving. It is used in the research to predict protein structures based on 112 known structures selected from the Brookhaven Protein Databank. The phi and psi angles of each amino acid in a protein are used to represent its 3-D structure. The authors extend MBR to include a recursive procedure to refine its initial prediction and a varying 'window' size to take into account the interaction between amino acids apart from different distances along the amino acid sequence. The system implemented, PHI-PSI, has been tested with all the available data. It does better than distribution-based guesses for most of the phi and psi angle values.