Deconvolution of Electron-Ionization Mass Spectra of Mixtures Without Chromatography

We explain a single electron-ionization mass spectrum as a sparse nonnegative combination of reference spectra from a library without any chromatographic separation. The result has practical value for compact volatile-organic-compound analysis workflows. The deconvolution of a spectrum into its constituent compounds is a challenging task: the support is discrete, parameters are continuous, and objectives are nonconvex in correlated libraries. On top of that, we handle small integer mass-to-charge shifts, moderate variability in relative fragment intensities, and instrument-induced noise. The proposed workflow first screens the library to produce a high-recall shortlist using ion-informed heuristics; then, a greedy builder adds one candidate at a time, with CMA-ES with margin and a local refinement used before evaluating the residual fitness. Candidate supports are scored with a replicate-calibrated likelihood of the observed spectrum under the reconstruction, augmented with penalties that discourage structured residuals and peak-shaped mismatch. Experiments on real spectra and synthetic mixtures show effective identification with practical runtime.