Collision-activated dissociation (CAD) of tryptic peptides is a cornerstone of mass spectrometrybased proteomics research. Principal component analysis of a database containing 15,000 high-resolution CAD mass spectra of gas-phase tryptic peptide dications revealed that they fall into two classes with a good separation between the classes. The main factor determining the class identity is the relative abundance of the peptide bond cleavage after the first two N-terminal residues. A possible scenario explaining this bifurcation involves trans-to cis-isomerization of the N-terminal peptide bond, which facilitates solvation of the N-terminal charge on the second backbone amide and formation of stable b 2 ions in the form of protonated diketopiperazines. Evidence supporting this scenario is derived from statistical analysis of the high-resolution CAD MS/MS database. It includes the observation of the strong deficit of a 3 ions and anomalous amino acid preferences for b 2 ion formation. do not predict relative abundances of fragment ions and are mostly concerned with matching ionic masses of certain fragment types. At the same time, fragment ion abundances in CAD MS/MS spectra can be relatively reproducible and are far from homogeneous-and to use the patterns of these abundances for sequence verification has for a long time been a tempting idea. Two alternative approaches were pursued in implementation of this idea. One approach was to collect a massive database of tandem mass spectra of all possible peptides and to make direct comparison between the standard MS/MS datasets and the experimental ones [3]. Although this approach has many merits, it faces a huge challenge in the form of an astronomical number of peptide sequences existing in nature. The alternative approach is to predict fragment abundances for a given sequence. The abundancepredicting algorithm by Zhang has more than 200 adjustable parameters but still exhibits only 70% accuracy [4,5]. The global model of CAD fragmentation developed by us for tryptic peptide dications uses only 20 parameters, one for each amino acid [6]. In preliminary assessment of the prediction accuracy of that model we have found a bifurcating behavior. It appeared that, although a large group of mass spectra behaved in good agreement with the model's predictions, another large group showed significantly deviating features. Thus the question arose as to whether all tryptic peptide dications constitute one group in the statistical sense. This question is critical for every abundance-predicting algorithm because different models of fragmentation should be applied to describe the behavior of each group. The obvious hypothesis that one group represented peptides ending with Lys and another one with Arg was quickly tested and rejected. Another suggestion was that the division line was the presence or absence of a particular amino acid, e.g., proline, histidine, or aspartic acid, which are known for their effect on bond fragmentation [7,8]. This hypothesis was also tested and rejected. An...