Convergent Fragmentation: A Small, Shared Ion Vocabulary Underlies the Limits of Structure–Spectrum Fidelity
Year: 2026
Type: preprint
Abstract: Tandem mass spectral similarity is widely used to infer whether molecules are structurally related, yet the fragmentation processes underlying this assumption remain incompletely understood. Structure–Spectrum Fidelity (SSF) measures the strength of this correspondence. Across structurally stratified compound pairs from the METLIN reference library, structural and spectral similarity were only moderately related (R2 = 0.28): closely related molecules could produce dissimilar spectra, while structurally different molecules could produce similar spectra. We investigated two fragmentation processes that could contribute to these opposing outcomes. First, differences in net hydrogen redistribution were associated with reduced spectral similarity between compound pairs. At corpus scale, hydrogen redistribution accounted for approximately 25–30% of model-explained fragment-ion intensity, equivalent to approximately 6.7% of total fragment-ion intensity. Second, chemically diverse precursors frequently converged onto the same recurrent product ions. Sharing an individual recurrent aromatic/allylic cation was uninformative, but elevated spectral similarity was concentrated among the relatively small subset of compound pairs sharing the greatest numbers of these ions. This was an upper-tail effect rather than a general increase with recurrent-ion sharing. Across 958,450 deduplicated METLIN compounds, we identified 56 recurrent product-ion formulas spanning five chemical families and both ionization modes; every family became more prevalent with increasing collision energy. Together, these findings identify complementary routes to limited SSF: compound-specific fragmentation can separate the spectra of related molecules, whereas extensive convergent fragmentation can make a subset of unrelated molecules appear spectrally similar. More broadly, across nearly one million molecules, fragmentation repeatedly converged on a shared vocabulary of only 56 product-ion formulas.
Source: ChemRxiv
Authors: Winnie Uritboonthai, Linh Hoang, Gary Siuzdak
Language: English
Subfield: Spectroscopy
Field: Chemistry
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7172521330