Learning-based event locating for single-molecule force spectroscopy

• Published in: Biochemical and biophysical research communications Vol. 556; pp. 59 - 64
• Main Authors: Lin, Zuzeng, Gao, Xiaoqing, Li, Shuai, Hu, Chunguang
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 04.06.2021
• Subjects: algorithms; Computational methods; data collection; Deep learning; humans; labor; neural networks; optical traps; Single-molecule spectroscopy; spectroscopy; Deep learning; Single-molecule spectroscopy; Computational methods
• ISSN: 0006-291X; 1090-2104; 1090-2104
• DOI: 10.1016/j.bbrc.2021.03.159

Acquiring events massively from single-molecule force spectroscopy (SMFS) experiments, which is crucial for revealing important biophysical information, is usually not straightforward. A significant amount of human labor is usually required to identify events in the measured force spectrum during measuring or before performing further data analysis. This prevents the experiment from being done in a fully-automated manner or scaling with the throughput of the measuring setup. In this work, we attempt to tackle this problem with a deep learning approach. A deep neural network model is developed to infer the occurrence of the events using the data stream from the measuring setup. We demonstrated that the proposed method could achieve high accuracy with force spectrums of a variety of samples from both optical tweezers and AFMs by learning from user-given samples instead of complicated manual algorithm designing or parameter tuning. Furthermore, we found that the trained model can be used to perform event detection on datasets measured from a different optical tweezer setup, showing the potential of being leveraged in more complex deep learning schemes. •This work proposed a user-friendly approach to massively locate the events of interest from experimental SMFS data streams.•To our best knowledge, this work is the first attempt to adopt deep learning methods to locate events in SMFS experiments.•It is demonstrated that the proposed method could learn to locate events for a variety of samples from different instruments including optical tweezers and AFMs.•This method could help to achieve automated measuring on SMFS high-throughput platforms or other deep-learning based SMFS signal analysis.