Network Topology Evaluation\nand Transitive Alignments\nfor Molecular Networking

Untargeted tandem\nmass spectrometry (MS/MS) is an essential technique\nin modern analytical chemistry, providing a comprehensive snapshot\nof chemical entities in complex samples and identifying unknowns through\ntheir fragmentation patterns. This high-throughput approach generates\nlarge data sets that can be challenging to interpret. Molecular Networks\n(MNs) have been developed as a computational tool to aid in the organization\nand visualization of complex chemical space in untargeted mass spectrometry\ndata, thereby supporting comprehensive data analysis and interpretation.\nMNs group related compounds with potentially similar structures from\nMS/MS data by calculating all pairwise MS/MS similarities and filtering\nthese connections to produce a MN. Such networks are instrumental\nin metabolomics for identifying novel metabolites, elucidating metabolic\npathways, and even discovering biomarkers for disease. While MS/MS\nsimilarity metrics have been explored in the literature, the influence\nof network topology approaches on MN construction remains unexplored.\nThis manuscript introduces metrics for evaluating MN construction,\nbenchmarks state-of-the-art approaches, and proposes the Transitive\nAlignments approach to improve MN construction. The Transitive Alignment\ntechnique leverages the MN topology to realign MS/MS spectra of related\ncompounds that differ by multiple structural modifications. Combining\nthis Transitive Alignments approach with pseudoclique finding, a method\nfor identifying highly connected groups of nodes in a network, resulted\nin more complete and higher-quality molecular families. Finally, we\nalso introduce a targeted network construction technique called induced\ntransitive alignments where we demonstrate effectiveness on a real\nworld natural product discovery application. We release this transitive\nalignment technique as a high-throughput workflow that can be used\nby the wider research community.