A3N: An artificial neural network n-gram-based method to approximate 3-D polypeptides structure prediction

A long standing problem in computational molecular biology is to determine the three-dimensional (3-D) structure of a protein when only a sequence of amino acids residues is given. Some protein structure prediction methods utilize structural information from protein templates in order to build the structure of unknown proteins. Examining structural protein motifs in detail is highly difficult since the task of mapping from a local sequence of amino acid residues to a local 3-D protein structure is very complex. This study presents a new statistical fragment-based method to acquire structural information from small protein template samples (A3N – Artificial Neural Network n-gram-based). Structural data obtained from protein templates were used in order to train an artificial neural network. Afterwards, approximative 3-D polypeptides structures are built through the use of a sequence-to-structure mapping function. The efficiency of the developed method is demonstrated in four case studies of polypeptides whose sizes vary from 19 to 34 amino acids residues. As indicated by the RMSD values and Ramachandran Plot values, the results show that the predicted structures adopt a fold similar to the experimental structures. Thus, they can be used as input structures in refinement methods based on molecular mechanics (MM), e.g. molecular dynamics (MD) simulations. The search space is expected to be greatly reduced and the ab initio methods can demand a much reduced computational time to achieve a more accurate polypeptide structure. We also discuss the results, future works and limitations of the proposed method.