Binary Coding, mRNA Information and Protein Structure

Abstract

We describe new binary algorithm for the prediction of α and β protein folding types from RNA, DNA and amino acid sequences. The method enables quick, simple and accurate prediction of α and β protein folds on a personal computer by means of a few binary patterns of coded amino acid and nucleotide physicochemical properties. The algorithm was tested with machine learning SMO (sequential minimal optimization) classifier for the support vector machines and classification trees, on a dataset of 140 dissimilar protein folds. Depending on the method of testing, the overall classification accuracy was 91.43% – 100% and the tenfold cross-validation result of the procedure was 83.57% – >90%. Genetic code randomization analysis based on 100,000 different codes tested for the protein fold prediction quality indicated that: a) there is a very low chance of p = 2.7 x 10^(-4) that a better code than the natural one specified by the binary coding algorithm is randomly produced, b)dipeptides represent basic protein units with respect to the natural genetic code defining of the secondary protein structure.