Compound identification via deep classification model for electron-ionization mass spectrometry
Compound identification in electron-ionization mass spectrometry (EI-MS) is usually achieved by matching the query mass spectrum to the well-collected reference spectral library. Although various similarity methods have been developed in recent years, it is still difficult to distinguish some simila...
Uložené v:
| Vydané v: | International journal of mass spectrometry Ročník 463; s. 116540 |
|---|---|
| Hlavní autori: | , , , |
| Médium: | Journal Article |
| Jazyk: | English |
| Vydavateľské údaje: |
Elsevier B.V
01.05.2021
|
| Predmet: | |
| ISSN: | 1387-3806, 1873-2798 |
| On-line prístup: | Získať plný text |
| Tagy: |
Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
|
| Shrnutí: | Compound identification in electron-ionization mass spectrometry (EI-MS) is usually achieved by matching the query mass spectrum to the well-collected reference spectral library. Although various similarity methods have been developed in recent years, it is still difficult to distinguish some similar mass spectra, especially for isomers. In this work, a deep learning based on classification model is proposed to improve the final compound identification. The replicate library of NIST05 is used as query data and main library is used as reference database. Through a simple library searching algorithm, a data set consisting of paired similar mass spectra is created. Based on the data set, two optimal deep classification models (model-1 and model-2) are trained to determine whether the query and candidate spectrum are same compound. To validate the proposed method, a series experiments are conducted by using the created classification model to aid random projection for compound identification. The experimental results show that the model-1 and model-2’s identification accuracy of Rank 1 are increased by 1.5% and 7.7% respectively. Contrasting three state-of-are similarity measure, the proposed classification model achieves the best identification performance.
Compound identification in EI-MS is usually achieved by matching the query mass spectrum to the reference spectral library. However, some similar mass spectra are difficult to distinguish by only using similar measure. In order to distinguish these similar mass spectra effectively, a deep learning based classification model is proposed to determine whether two spectra are same compound. By using random projection as library searching algorithm, a data set consisting of paired similar mass spectra is created to train the proposed classification model. The classification results is used to alter the TOP-N mass spectra rank. The experimental results show that the proposed model achieves the best identification performance. [Display omitted]
•A data set consisting paired similar mass spectra is created from NIST05 spectral library.•A novel deep classification model is proposed to aid compound identification.•The proposed model can improve the identification accuracy effectively. |
|---|---|
| ISSN: | 1387-3806 1873-2798 |
| DOI: | 10.1016/j.ijms.2021.116540 |