Cores extraction based neural network for proteins fold recognition, in "JOBIM’05

Abstract: Motivation: Identifying the fold class of a protein sequence of unknown structure is a fundamental problem in modern biology. In this context, we present an original approach for treating 3D structures of proteins. This approach consists on both, generating a significant data set for automatic learning and processing a modeling system for fold recognition. In the first step, the method consists to calculate a structural alignment for each class (family) of structures using the combinational extension (CE) alignment methodology. In the second step, we carry out from each alignment’s matrix, a taxonomy created with ascendant hierarchical clustering (AHC). Then, structural cores are extracted from all families and each core represents a prototype for each class of structures. Finally, a neural network system (NN) is built from a matricial modeling based sequence coding for protein fold recognition. Subsequently, with the help of cores extraction, a data set is generated to build a strong fold recognition system which represents more than 75 % classification accuracy over 21 CATH families