Predicting\nTandem Mass Spectra of Small Molecules\nUsing Graph Embedding of Precursor-Product Ion Pair Graph

Liquid chromatography–mass spectrometry (LC-MS)-based\nmetabolomics\nidentification relies heavily on high-quality MS/MS data; MS/MS prediction\nis a good way to address this issue. However, the accuracy of the\nprediction, resolution, and correlation with chemical structures have\nnot been well-solved. In this study, we have developed a MS/MS prediction\nmethod, PPGB-MS2, which transforms the MS/MS prediction into fragment\nintensity prediction, and the concept of precursor-product ion pair\ngraph bags (PPGBs) was introduced to represent fragments, achieving\nuniform representation of precursor and product ion structures and\nMS/MS fragmentation information. The chemical structure information\nis kept before it is incorporated into machine learning models. Due\nto the PPGB representation, graph neural networks (GNNs) can be utilized\nto achieve MS/MS fragment intensity prediction. The system was trained\nand evaluated using [M+H]+ and [M-H]- data acquired by an Agilent\nQTOF 6530 in the NIST 20 tandem MS database. Results demonstrated\nthat the average cosine similarity is 0.71 in the test set, which\nis higher than classical MS/MS prediction methods. PPGB-MS2 also achieves\nhigh-resolution MS/MS prediction due to its effective management of\nthe correspondence between fragments and structures.