SVM-based automatic classification for protein structural domain

The automatic classification for protein structure plays an important role in bioinformatics. Here we present an improved multiclass SVM for the classification based on the features of the protein structure which were extracted from the protein convex hull. Firstly, we modify the gauss radial kernel by adding a positive constant to the kernel function. Secondly, we take weighted SVM to deal with the imbalanced dataset. Experiments demonstrate the superiority of our new strategies. In addition, we design the hierarchical classifier which is more suitable to the CATH database.