A structure-aware platform for comprehensive lipidomics annotation and quantification

Current lipidomics pipelines struggle with extensive lipid structural diversity and limited coverage. We present an end-to-end intelligent analysis platform that integrates annotation and rapid quantitative analysis in a platform-independent software solution. It incorporates a mixture-of-experts (MoE) model that jointly learns mass spectral information and lipid structural features to enable structure-aware spectral interpretation, Crucially, it uniquely achieves seamless compatibility with both electron-activated dissociation (EAD) and Paternò–Büchi (PB) derivatization spectra—a capability not offered by any existing platform—alongside conventional MS/MS. It also features a manually curated 972.1-million-entry dynamic lipid fragmentation library, designed to maximize structural coverage across conventional and emerging lipid space by incorporating diverse fatty-acyl chain acylation patterns, sulfated and amino-acid-conjugated bile acids, lipid A species, glycolipids, and emerging lipid classes. The platform delivers ultra-fast spectral searching and high annotation accuracy (MedRE <0.3%, FDR 5.7%). The platform also features a rapid, accurate, and reproducible quantification module, enabling streamlined lipid quantitation across diverse acquisition modes and experimental designs. Leveraging its unique PB compatibility, we identified lipid isomer markers that robustly distinguish breast cancer, benign breast nodules, breast cancer lung metastasis, and primary lung cancer—discrimination unattainable by other methods due to their inability to jointly process PB and EAD spectra. This platform significantly advances comprehensive lipidomics analysis.