Prediction of mutation positions in H5N1 neuraminidases from influenza A virus by means of neural network.

Quantification of mutation capacity within a protein could be a way to model the mutation relationship not only because history might not leave many cues on the causes for mutations but also the evolved protein might no longer be subject to previous mutation causes. Randomness should play a constant role in engineering mutations in proteins because randomness suggests the maximal probability of occurrence by which a protein would be constructed with the least time and energy to meet the speed of rapidly changing environments. Since 1999, we have developed three approaches for quantifying of randomness of protein by which each amino acid has three numeric values. In this study, we model our three random numeric values in each amino acid with occurrence and non-occurrence of mutation, which are classified as unity and zero, using a 3-6-1 feedforward backpropagation neural network to predict the mutation positions in H5N1 neuraminidases. The results show that the neural network can capture the mutation relationship as measured by prediction sensitivity, specificity, and total correct rate. With the help of translation probability between RNA codes and mutated amino acids, we predict the would-be-mutated amino acids at predicted mutation positions.