| 研究開始時の研究の概要 |
Mass spectrometry is a common technique in analytical chemistry for metabolite identification. The aim of our reseach is to develop advanced machine learning models to identify metabolites from their mass spectra with the following main criteria: 1) High accuracy: given a query spectrum, the proposed models are expected to produce a highly accurate list of candidate with most similar spectra; 2) Computational efficiency: in order to process large-scale datasets of metabolites in reality such as PubChem, it is desirable for the proposed models to produce candidates with fast prediction.
|
| 研究実績の概要 |
The results of this research are summarized in four publications. The first publication is a review paper on computational methods for metabolite identification from mass spectra, which appeared in Briefings in Bioinformatics. In the second publication, we proposed machine learning models with the incorporation of peak interactions for predicting molecular fingerprints, which appeared in Bioinformatics (also the proceedings of the 26th International Conference on Intelligent Systems for Molecular Biology (ISMB)). In the third publication, we proposed another machine learning model for predicting representations for metabolites from their chemical structures and spectra, which appeared in Bioinformatics (also the proceedings of the 27th ISMB). All together, we summarized recent advances of machine learning for metabolite identification from mass spectra, including our research results, in a book chapter, which will appear in Creative Complex Systems, Springer 2021.
|
