Predicting Protein Structural Class for Low-Similarity Sequences via Novel Evolutionary Modes of PseAAC and Recursive Feature Elimination