정영흔 (영남대학교 약학대학)

김주현 (영남대학교 약학대학)

정영흔 (영남대학교 약학대학)

장예나 (영남대학교 약학대학)

김주현 (영남대학교 약학대학)

In untargeted lipidomics, accurate computational annotation remains a primary bottleneck, often requiring time-consuming manual curation. To evaluate lipidomic analytical tool performance, we systematically compared the annotation capabilities and coverage of MZmine, LipiDex, and GNPS2 using a HepG2 lipidomic dataset acquired via RPLC and HILIC in both positive (+) and negative (−) ionization modes. Comparative analysis revealed that cosine similarity-based annotation via GNPS2 risks feature over-interpretation. Conversely, MZmine and LipiDex, tools containing rule-based in silico lipid fragment libraries, provided more reliable species- or molecular species-level annotations grounded in experimental spectral evidence. While many annotated features overlapped between MZmine and LipiDex, a substantial fraction was exclusively annotated by a single platform. Furthermore, cross-platform comparison among RPLC (±) and HILIC (±) modes demonstrated high proportions of method-specific lipids, with RPLC (+) achieving the highest overall annotation rate. In conclusion, utilizing specialized, rule-based software is critical to preventing misannotation, though coverage varies by data acquisition methods. Ultimately, integrating both RPLC and HILIC in dual-polarity modes is suggested for maximizing lipidome coverage in untargeted lipidomics workflows.

​