Folding rate prediction based on neural network model

Abstract

One of the most important challenges in biology is to understand the relationship between the folded structure of a protein and its primary amino acid sequence. A related and challenging task is to understand the relationship between sequences and folding rates of proteins. Previous studies found that one of contact order (CO), long-range order (LRO), and total contact distance (TCD) has a significant correlation with folding rate of protein. Although the predicted results from TCD can provide better results, the deviation is also large for some proteins. In this paper, we adopt back-propagation neural network to study the relationship between folding rate and protein structure. In our model, the input nodes are CO, LRO, and TCD, and the output node is folding rate. The number of nodes in the hidden layer is seven. Our results show that the relative errors for the predicted results are even lower than other methods in the literature. We also observe a best excellent correlation between the folding rate and contact parameters (including CO, LRO, and TCD), and find that the folding rate depends on CO, LRO and TCD simultaneously. This means that CO, LRO and TCD are similarly important in folding rate of protein. Some comparisons are made with other methods.