Abstract

Poster Presentations

Day 3, May 17（Wed.）　　Room P (Foyer, Room 1004-1007)

3P-50　　PDF

Development of predicted mass spectrum for electron ionization method by machine learning and qualitative analysis method

(JEOL)
^oKenji Nagatomo, Ayumi Kubo, Azusa Kubota, Masaaki Ubukata, Haruo Iwabuchi

The electron ionization (EI) method is widely used as an ionization method for gas chromatograph mass spectrometry (GC/MS). Fragment ions are mainly observed in a mass spectrum obtained by the EI method. Because fragment ions reflect the structure of a compound, the compounds can be identified by comparing with the mass spectrum database. About 300,000 mass spectra are registered in NIST20, the most widely used database for GC/MS. However, as of 2023, more than 100 million compounds are registered in PubChem, a general compound database that does not hold mass spectra. In other words, most of the compounds that do not have mass spectra, so the compounds cannot be identified even by database searches. In this study, we used machine learning to create a model for predicting mass spectra from structural formulas, and constructed a database that includes predicted mass spectra for PubChem's compounds. A new qualitative analysis method incorporating this database was developed. In this presentation, we will report the accuracy of predicted mass spectra and the application of the developed qualitative analysis method with this database to unknown compounds.