Protein Folding Rate Prediction Based On Neural Network Model

The protein is a type of macromolecules that occupies a special position in the living-beings. Protein folding is the process by which a protein progress from its denatured state to its specific biologically native conformation. In this process, the folding rates of different proteins are different largely.In order to reveal the determinant factors of folding rate, many researchers made a great deal of research work. By far, many empirical parameters based on native structural information have been proposed successively, such as contact order, long-range order, secondary structure content, effective length, etc. The predicted results by the methods based on these parameters fit well with the experimental folding rates.However, many of the above-mentioned methods need to know protein structure information, or to predict them in advance, for the purpose of predicting a protein's folding rate. In the present work, we proposed a model using artificial neural network to predict protein folding rate directly from amino acid sequence without any structure information needed. Firstly, convert the amino acid sequence to numeric sequence according to its biological characteristics, such as its average free energy in helical conformation or its frequency in sequence; Secondly, use wavelet analysis method which is widely adopted in modern digital signal processing to process the boundary of numerical sequence, such as symmetry extension, zero-padding extension and smooth extension, then use sliding Window technique based on Gauss functions to compress the data; finally establish a three-layer BP model using artificial neural network toolbox of matlab to predict protein folding rate. The correlation coefficient is 0.635 in jackknife cross validation tests, which is comparable with the results of the past methods. The establishment of this model opens a new way for protein folding rate.