Identification of diagnostic reporters in MS/MS Molecular Networks

Abstract Untargeted metabolomics frequently utilizes molecular networking (MN) to map mass spectrometry data, enhancing compound annotation workflows. However, identifying which fragmentation features (fragment ions/neutral losses) can be considered characteristics of a molecular families rather than representing generic, non-specific fragments, remains a major challenge. Here, we report a data-driven framework that facilitates MN interpretation by highlighting the fragmentation features that act as signatures for MN subnetworks. Specifically, we apply a network modularity approach coupled with Quantile Generalised Additive Models (QGAM) to identify fragment ions and neutral losses shared across nodes that exhibit higher topological cohesiveness than expected for their level of occurrence. By isolating the chemical signatures typical of an MN region, we provide an objective foundation for annotation propagation. This approach successfully prioritized class-specific diagnostic reporters in the Alternaria metabolome, distinguishing ubiquitous fragmentation patterns from diagnostic tricycloalternarene reporters. Our method provides a data-dependent strategy to guide annotation propagation and accelerate structural elucidation in complex untargeted datasets.