Analyzing protein dynamics from fluorescence intensity traces using unsupervised deep learning network

• Published in: Communications biology 2020-11, Vol.3 (1), p.669-669, Article 669
• Main Authors: Yuan, Jinghe, Zhao, Rong, Xu, Jiachao, Cheng, Ming, Qin, Zidi, Kou, Xiaolong, Fang, Xiaohong
• Format: Article
• Language: English
• Subjects: 631/1647/794; 631/337/2265; 631/45/56; 96/34; 96/35; 96/63; 96/95; Biology; Biomedical and Life Sciences; Carbocyanines - analysis; Carbocyanines - metabolism; Deep Learning; Diffusion; Fluorescence; Fluorescent Dyes - analysis; Fluorescent Dyes - metabolism; HeLa Cells; Humans; Information processing; Labeling; Life Sciences; Long short-term memory; MCF-7 Cells; Membrane proteins; Membrane Proteins - analysis; Membrane Proteins - chemistry; Membrane Proteins - metabolism; Microscopy, Fluorescence; Unsupervised Machine Learning
• ISSN: 2399-3642; 2399-3642
• DOI: 10.1038/s42003-020-01389-z

We propose an unsupervised deep learning network to analyze the dynamics of membrane proteins from the fluorescence intensity traces. This system was trained in an unsupervised manner with the raw experimental time traces and synthesized ones, so neither predefined state number nor pre-labelling were required. With the bidirectional Long Short-Term Memory (biLSTM) networks as the hidden layers, both the past and future context can be used fully to improve the prediction results and can even extract information from the noise distribution. The method was validated with the synthetic dataset and the experimental dataset of monomeric fluorophore Cy5, and then applied to extract the membrane protein interaction dynamics from experimental data successfully. Yuan et al. propose an unsupervised deep learning network approach to analyze the dynamics of membrane proteins from the fluorescence intensity traces. The unsupervised nature facilitates training of the system without predefined state number or pre-labelling and can even extract information from noise distribution.