Prediction of protein structural classes for low-homology sequences based on predicted secondary structure