On de novo interpretation of tandem mass spectra for peptide identification