Summary form only given. Prediction of protein structures from protein sequences using computers is an important step to discover proteins' 3D conformation structures and their functions and hence has profound theoretical and practical significance in areas such as protein engineering and drug design. In this paper, we discuss our new results in protein secondary structure and transmembrane protein prediction using support vector machines. We also discuss how to use a combination of support vector machine and decision tree to understand how a prediction is reached through rule extraction. Clearly, a good interpretation is useful for guiding biological experiments and may lead further prediction improvement. A novel approach of rule clustering for super-rule generation is also briefly discussed