Automatic prediction of protein domains from sequence information using a hybrid learning system