LEARNING PROTEIN SECONDARY STRUCTURE FROM SEQUENTIAL AND RELATIONAL DATA