Table I Sensitivity of selecting the correct pair of peptides from the spectral library and protein sequence database. MixDB and M-SPLIT were evaluated on a set of 5000 simulated mixture spectra. Each row indicates the percentage of cases when the top ranking pair is correct; numbers in parenthesis indicate fraction of cases when one of the top ten peptide pairs is correct. Two types of database search were performed. In the first, each spectrum was searched against all Yeast tryptic peptides. In the second, each spectrum was searched only against peptides with spectra in the NIST Yeast spectral library. In general M-SPLIT spectral library search was found to have better performance than MixDB database search. However, as shown in the fourth column, the main advantage of spectral library search was the reduction in the number of possible peptide candidates. We also compared the MixDB's performance with an iterative strategy where: (1) the dominant peptide is identified; (2) peaks explained by the top peptide match are removed from the spectrum, and (3) the residual spectrum is again searched against the database to identify the second peptide. When the mixture coefficient α is high the iterative strategy has similar performance to MixDB; however, as α becomes smaller, MixDB's scoring of both peptides at the same time results in considerably better performance than the iterative strategy
Mixture coefficient (α)M-SPLITMixDB (whole Yeast database)MixDB (library peptides only)Iterative method
1:19787 (97)95 (98)81
1:0.59279 (92)90 (98)74
1:0.38066 (86)79 (92)57
1:0.26350 (77)69 (87)30
1:0.13419 (43)34 (70)6