Poster Presentations
- Day 2, May 12（Tue.） Poster (1008/09)
-
2P-37（2B-O5-0910） PDF
Deep Learning for Identification of Organic Molecules from a Database of Mass Spectrometry and Visualization of the Attribution of Substructure to Peaks
The identification of a molecular structure of a sample from its mass spectra is still a challenging task. A rule-based or machine learning methods are widely studied for the identification. In the rule-based methods, it is sometimes difficult to prescribe the rules of complex ionization. In the machine learning methods, the identification from the liquid chromatography followed by tandem mass spectra (LC-ESI-MS/MS) for metabolites show high accuracy but there is not much research on the identification from gas chromatography followed by electron ionization mass spectra (GC-EI-MS). We developed a new method by deep learning for the identification of the molecular structure of the sample from EI-MS. In this method, to prescribe a rule or feature term by hand is not required. Because of this, our methods can be applied to a variety of molecules. We demonstrate that our method can generate sample molecular structure from GC-EI-MS. And then, we demonstrate to visualize the relationship between the substructure of the generated molecular structure and the peak in the query spectra. The result of this visualization shows that our method recognized fragmentation ion by the peak in the spectra and built a molecular structure from structures of fragmentation ion.