MS$^2$-Transformer: An End-to-End Model for MS/MS-assisted Molecule Identification

Mass spectrometry (MS) acts as an important technique for measuring the mass-to-charge ratios of ions and identifying the chemical structures of unknown metabolites. Practically, tandem mass spectrometry (MS/MS), which couples multiple standard MS in series and outputs fine-grained spectrum with fragmental information, has been popularly used. Manually interpreting the MS/MS spectrum into the molecules (i.e., the simplified molecular-input line-entry system, SMILES) is often costly and cumbersome, mainly due to the synthesis and labeling of isotopes and the requirement of expert knowledge. In this work, we regard molecule identification as a spectrum-to-sequence conversion problem and propose an end-to-end model, called MS$^2$-Transformer, to address this task. The chemical knowledge, defined through a fragmentation tree from the MS/MS spectrum, is incorporated into MS$^2$-Transformer. Our method achieves state-of-the-art results on two widely used benchmarks in molecule identification. To our best knowledge, MS$^2$-Transformer is the first machine learning model that can accurately identify the structures (e.g., molecular graph) from experimental MS/MS rather than chemical formula/categories only (e.g., C$_6$H$_{12}$O$_6$/organic compound), demonstrating it the great application potential in biomedical studies.