Neural network based prediction of 3D protein structure as a function of enzyme family type and amino acid sequences

Prediction of dihedral angles from amino acid sequences based on the neural network to predict protein structure is promising in the field of bioinformatics. the present proposed study presents a prediction tool for 3-dimensional (3d) protein structure as a function of enzyme family types and amino acid sequences. 11 different families of enzymes were investigated amounting to 97 enzymes in total. correlation of sequence with geometry coordinates as a function of amino acid descriptors and family class were generated through a neural network to predict coordinates. the structural-features of dissected triplets show significant influence on family type. r-values for the inter-family dataset as well as feature selection were not satisfying. in contrast,r-values around 0.8 were achieved in the case of intra-family prediction. furthermore,about 55 % of features were eliminated with a limited negative influence of 13% on the r-value. we believe that the present study provides a promising prediction method that advance computational methods in bioinformatics,especially to predict 3d protein structure as a function of enzyme family type and amino acid sequences. however,intra-family prediction probability is higher while using only one type of analysis based on the dihedral angles of turn structures of enzyme families. © 2017 jordan journal of biological sciences.