Figure 4. Comparison of different decoy strategies. The estimated FDR of each decoy method is compared with the real FDR. “Decoy: + 11” means increasing the mass of each Y ion by 11 Da, and “Decoy: + [1, 30]” means increasing the mass of each Y ion by a random mass ranging from 1–30 Da. “Decoy: Reverse” is simply reversing the Y ions. The finite mixture model is employed for all these decoy methods. “Decoy: + [1, 30]” is the closest estimation to the “Real FDR”.

Feasibility of the spectrum-based decoy method. In pGlyco, the spectrum-based decoy strategy coupled with the finite mixture model was used to estimate the glycan FDR. This novel strategy had been tested in routine peptide identification problems (see Figs S-3 and S-4 in the Supporting Information), and it was still necessary to test if it actually worked in a real glycopeptide dataset. To further validate the spectrum-based decoy method, we manually checked all the GPSM results. At 1% glycan FDR, 1 out of 556 GPSMs was confirmed as the false identification, while before glycan FDR cutoff, 73 GPSMs out of 765 GPSMs were confirmed as false identifications. The real error rate was calculated based on manually checked results. Although using the 2,704 GSMs to validate the glycan FDR estimation was more rational, it was difficult to judge if a GSM was correct without the identity of the peptide backbone. So here we manually checked the 765 GPSMs, since the peptide backbones were identified at 1% peptide FDR, which meant, when an incorrect GPSM occurred, it was probably because of the incorrect identification of glycan. The real FDR and reported FDR were compared, as shown in Fig. 4. When the reported FDR (the orange curve) was 1%, the real FDR (the cyan curve) was only 0.18%, which showed a conservative estimation by the spectrum-based decoy method. Based on our manually checked results, three spectrum-based decoy strategies were compared: increasing the mass of each Y ion by a fixed 11 Da30, or increasing the mass of each Y ion by a random mass ranging from 1 to 30 Da (our decoy method), or reversing the Y ions of each glycan (we defined the reversed Y ion mass = entire glycan mass – Y ion mass), and the results were also shown in Fig. 4. When increasing the Y ion masses by 11 Da as decoys, the FDR was far overestimated. And the FDR was also far overestimated by reversing the Y ions of each glycan structure as decoys, the reason might be that some reversed Y ions of a glycan structure were the same as target Y ions. For example, Figure S-5 in the Supporting Information showed that the Y ions of a glycan structure with the composition (3, 4, 0, 0, 0) were approximately identical to their reversals. Therefore, through our investigation in Fig. 4, Figs S-3 and S-4 in the Supporting Information, our spectrum-based decoy method was proved to be a fine choice to estimate the glycan FDR. And based on our manually checked data and the novel FDR estimation method, we also investigated the best condition for the Y1 ion filtration, as shown in the section “Best parameter for the Y1 ion filtration” and Table S-3 in the Supporting Information.

Figure 5. Fragmentation behavior of trimannosyl core and non-trimannosyl core ions in HCD-

(@NCE = 40%) and CID-MS/MS. (a) The number of trimannosyl core ions produced by HCD-MS/MS, or CID-MS/MS, or “HCD + CID” (combination of these two kinds of spectra). (b) The number of each trimannosyl core ion uniquely observed in one type of MS/MS. (Y-00000 is the Y0 ion and Y1_-00000 is the Y1_ ion. The core-fucosylated ions, Y-01001 and Y-02001, are also counted as trimannosyl core ions). (c) The number of non-trimannosyl core ions produced by HCD- and CID-MS/MS. In (c) there are 306 CID-MS/MS spectra not shown due to more than 9 non-trimannosyl core ions matched.