Northwestern Argonne Institute of Science and Engineering, Northwestern University, 2205 Tech Drive Suite 1-160, Evanston, 60208 Illinois, United States.
Mathematics and Computer Science Division, Argonne National Laboratory, 9700 S. Cass Avenue, Lemont, 60439 Illinois, United States.
ACS Sens. 2022 Sep 23;7(9):2710-2720. doi: 10.1021/acssensors.2c01218. Epub 2022 Aug 30.
Pulse-like signals are ubiquitous in the field of single molecule analysis, , electrical or optical pulses caused by analyte translocations in nanopores. The primary challenge in processing pulse-like signals is to capture the pulses in noisy backgrounds, but current methods are subjectively based on a user-defined threshold for pulse recognition. Here, we propose a generalized machine-learning based method, named pulse detection transformer (PETR), for pulse detection. PETR determines the start and end time points of individual pulses, thereby singling out pulse segments in a time-sequential trace. It is objective without needing to specify any threshold. It provides a generalized interface for downstream algorithms for specific application scenarios. PETR is validated using both simulated and experimental nanopore translocation data. It returns a competitive performance in detecting pulses through assessing them with several standard metrics. Finally, the generalization nature of the PETR output is demonstrated using two representative algorithms for feature extraction.
在单分子分析领域，脉冲信号无处不在，包括由纳米孔中分析物转位引起的电脉冲或光脉冲。处理脉冲信号的主要挑战是在噪声背景中捕获脉冲，但目前的方法主要基于用户定义的脉冲识别阈值。在这里，我们提出了一种基于广义机器学习的方法，称为脉冲检测变压器（Pulse Detection Transformer，简称 PETR），用于脉冲检测。PETR 确定单个脉冲的起始和结束时间点，从而在时间序列迹中单独提取脉冲段。它是客观的，不需要指定任何阈值。它为特定应用场景的下游算法提供了通用接口。我们使用模拟和实验纳米孔转位数据对 PETR 进行了验证。通过使用几个标准指标对其进行评估，它在检测脉冲方面表现出了有竞争力的性能。最后，使用两种用于特征提取的代表性算法证明了 PETR 输出的泛化性质。
