Park Ji In, Kim Myeong Ji, Lee Kyu Hyeong, Oh Seung Hyun, Kang Young Hoon, Kim Hyunwoo
College of Pharmacy and Integrated Research Institute for Drug Development, Dongguk University-Seoul, Goyang 10326, Republic of Korea.
Plants (Basel). 2024 Dec 4;13(23):3401. doi: 10.3390/plants13233401.
A vision transformer (ViT)-based deep neural network was applied to classify the flavonoid glycoside isomers by analyzing electrospray ionization tandem mass spectrometry (ESI-MS/MS) spectra. Our model successfully classified the flavonoid isomers with various substitution patterns (3-O, 6-C, 7-O, 8-C, 4'-O) and multiple glycosides, achieving over 80% accuracy during training. In addition, the experimental spectra from flavonoid glycoside standards were acquired with different adducts, and our model showed robust performance regardless of the experimental conditions. As a result, the vision transformer-based computer vision model is promising for analyzing mass spectrometry data.
一种基于视觉Transformer（ViT）的深度神经网络被应用于通过分析电喷雾电离串联质谱（ESI-MS/MS）光谱来对黄酮糖苷异构体进行分类。我们的模型成功地对具有各种取代模式（3-O、6-C、7-O、8-C、4'-O）和多种糖苷的黄酮异构体进行了分类，在训练过程中准确率超过80%。此外，还获取了来自黄酮糖苷标准品的具有不同加合物的实验光谱，并且我们的模型在无论实验条件如何的情况下都表现出稳健的性能。因此，基于视觉Transformer的计算机视觉模型在分析质谱数据方面具有广阔前景。
