Evaluating Training Procedures in a Naive Bayes Approach to Pathogenicity Prediction: The Case of the Family of Sodium Channel Proteins

Abstract

PolyPhen-2 is a software that could help predict the pathogeneicity of mutations. We used it for prediction for sodium channel protein data after comparison with a few other prediction including Grantham Scores, SIFT, phyloP, PolyPhen-2, and CADD. But it’s still working with limited accu- racy. We are primarily concerned about is that our problem is trying to study the pathogenicity of sodium channel proteins and we are question- ing that if these softwares will help us predicting these mutation as well as the training set is based on the whole genome. So we tried to modify the software by training a data set from sodium channel protein mutations using naive Bayes algorithms. Then we compared the prediction results from our classifier and the prediction results of PolyPhen-2, then we could see that the classifier trained by our methods can make prediction results with significantly better accuracy.