Prediction of protein tertiary structure via the “translation” of amino acid sequences into internal torsional angles.
Proteins are abundant in virtually all biological processes within the cells of, not just humans, but all living organisms. Sequences of amino acids undergo a spontaneous transition into their final 3-dimensional structure, called folding. Structure determination is important as the structure dictates the function of the protein. This report devises a novel method of protein structure prediction that draws on recent successes in the fields of Neural Machine Translation and Natural Language Processing, applying them to the field of Bioinformatics. The proposed method is an attentive encoder-decoder network that translates embedded ngrams of amino acid sequences into dihedral angles, feeding them into the PNeRF algorithm for a cartesian representation. The results are then benchmarked against the contestants of CASP13.