Structure-guided parameter optimization of feature-based molecular networking for DIA-HRMS screening of antihistamines in cosmetics.

Feature-based molecular networking (FBMN) can support prioritization of library-absent features in high-resolution mass spectrometric screening, but parameter choices may create chemically misleading links, particularly for data-independent acquisition (DIA) spectra. We developed a structure-guided multi-metric parameter-optimization framework using 51 authenticated antihistamines assigned to structural classes independently of retention time, MS/MS similarity, and network topology. Across 595 GNPS-FBMN parameter combinations, the selected application-specific setting P264 (minimum cosine similarity, 0.475; matched ions, 8; MAX_SHIFT, 150 Da) gave an adjusted Rand index of 0.216, a same-class edge ratio of 0.697, and a cross-class edge ratio of 0.303, compared with 0.101, 0.621, and 0.379, respectively, for the controlled GNPS edge-default setting. In 120 commercial cosmetics, in-house library screening detected diphenhydramine in one sample; authentic-standard comparison confirmed the retention time, precursor ion, and MS/MS evidence, and post-confirmation measurement gave 3.28 mg kg-1. Independent 51-target triple-quadrupole MRM analysis produced no discordant known-target result in the evaluated commercial-sample set. In a separate library-absent proof-of-concept experiment, buclizine fortified into blank cream satisfied the predefined three-anchor prioritization rule, ranked first with SIRIUS/CSI:FingerID, and was subsequently confirmed with an authentic standard. These results support the optimized FBMN as a complementary approach for structural-context assessment and library-absent candidate prioritization under the evaluated DIA-HRMS conditions.