Protein three-dimensional structural forecasting manner

摘要

PURPOSE:To predict each local structure of the three-dimensional structure of a structure-unknown protein from amino acid sequence information on the protein with high precision. CONSTITUTION:In a step 21, structure-known and structure-unknown protein amino acid sequence are inputted and plus examples of local structure areas are outputted from their alignment (matching). In a step 22, protein amino acid sequence whose three-dimensional structures are already known are inputted and negative examples of local structure areas are outputted from their alignment. In a step 23, the plus examples and negative examples are inputted and estimated values of real numeral parameters of a probable rule are calculated from the amino acid kind or real numeral attributes of training data. In a step 24, the structure of the probable rule representing the dependency relation among variables corresponding to respective residual radical positions is determined on the basis of information amount standards. In a step 25, a test data sequence is inputted and the stochastic rule is used to output activity indicating the probability that the respective areas of the test data sequence are in local structure. In a step 26, activity values are inputted and an optimum value among them is outputted.