[8a-N302-1~11] 18.1 Materials Informatics

10:45 AM - 11:00 AM JST(1:45 AM - 2:00 AM UTC)

[8a-N302-7] Forward Prediction of LC-MS/MS Spectra Using a Chemical Language Model and Stepwise Learning

〇(B)Aoi Takahashi¹, Satoki Muto¹, Takashi Fujii¹, Takahiro Umemoto¹, Akiko Kumada¹, Masahiro Sato¹ (1.Tokyo Univ.)

Keywords:

mass spectra prediction,transfer learning

In compound identification by LC-MS/MS, the lack of reference spectra is a major challenge. In this study, we worked on constructing a forward model that predicts mass spectra from chemical structures. Based on a pre-trained chemical language model, we performed pre-training with computed spectra, followed by stepwise transfer learning using a public database and spectra acquired on the target measurement system. With measurement conditions given as input, we evaluated how the learning at each stage affects prediction accuracy. We report the prediction performance and its limiting factors.