Quantifying Fragment Ambiguity in LC-HRMS/MS Spectral Databases

LC–HRMS/MS-based non-target analysis relies heavily on MS2 spectral matching for compound annotation. However, not all fragments contribute equally to identification confidence, and quantitative measures of fragment ambiguity remain lacking. In this study, we developed a data-driven framework to assess fragment-level ambiguity through the Fragment Ambiguity Score (FAS), which combines fragment stability and specificity within precursor-mass-defined candidate spaces. FAS was evaluated using five experimental spectral databases (MassBank, MoNA, GNPS, NIST20, and MSnLib), comprising together 1,026,855 spectra and an in silico spectral library containing 256,865 spectra. The results showed that MS2 spectra contain both diagnostic and ambiguous fragments, with diagnostic fragments representing only a subset of the total fragmentation information available for a compound. Only 14.1% of fragments in the experimental databases and 10.7% in the in silico library exhibited low ambiguity (FAS ≤ 0.5). Fragment ambiguity showed a strong correlation with database composition, with fragments appearing highly diagnostic in smaller databases often becoming substantially more ambiguous when evaluated against larger and more diverse spectral collections. Differences between experimental and in silico databases further demonstrated the influence of database origin and composition on fragment ambiguity estimates. These findings demonstrate that the number of matched fragments alone is insufficient for confidence assessment and highlight the importance of complementary sources of evidence. The proposed framework provides a quantitative basis for assessing fragment ambiguity in spectral databases.