Analyzing the locomotory gaitprint of Caenorhabditis elegans on the basis of empirical mode decomposition

The locomotory gait analysis of the microswimmer, Caenorhabditis elegans, is a commonly adopted approach for strain recognition and examination of phenotypic defects. Gait is also a visible behavioral expression of worms under external stimuli. This study developed an adaptive data analysis method based on empirical mode decomposition (EMD) to reveal the biological cues behind intricate motion. The method was used to classify the strains of worms according to their gaitprints (i.e., phenotypic traits of locomotion). First, a norm of the locomotory pattern was created from the worm of interest. The body curvature of the worm was decomposed into four intrinsic mode functions (IMFs). A radar chart showing correlations between the predefined database and measured worm was then obtained by dividing each IMF into three parts, namely, head, mid-body, and tail. A comprehensive resemblance score was estimated after k-means clustering. Simulated data that use sinusoidal waves were generated to assess the feasibility of the algorithm. Results suggested that temporal frequency is the major factor in the process. In practice, five worm strains, including wild-type N2, TJ356 (zIs356), CL2070 (dvIs70), CB0061 (dpy-5), and CL2120 (dvIs14), were investigated. The overall classification accuracy of the gaitprint analyses of all the strains reached nearly 89%. The method can also be extended to classify some motor neuron-related locomotory defects of C. elegans in the same fashion.秀丽隐杆线虫（Caenorhabditis elegans）的运动步态分析，是实现线虫品系识别与表型缺陷检测的常用研究手段。步态亦是线虫在外界刺激下的直观行为表现形式。本研究构建了一种基于经验模态分解（empirical mode decomposition, EMD）的自适应数据分析方法，用以揭示复杂运动背后的生物学线索。该方法可根据线虫的步态特征（即运动表型特征）对其品系进行分类。首先，从目标线虫中构建运动模式的基准特征。将线虫的身体曲率分解为四个固有模态函数（intrinsic mode functions, IMFs）。随后，将每个固有模态函数划分为头部、躯干中段与尾部三个区段，以此生成可展示预设数据库与实测线虫间相关性的雷达图。通过k均值聚类（k-means clustering）计算得到综合相似度得分。本研究还生成了基于正弦波的模拟数据，以评估所提算法的可行性。实验结果表明，时间频率是该分析流程中的主要影响因素。实际研究中共对五种线虫品系展开了考察，分别为野生型N2、TJ356（zIs356）、CL2070（dvIs70）、CB0061（dpy-5）以及CL2120（dvIs14）。所有品系的步态特征分析总体分类准确率接近89%。该方法还可通过相同范式拓展应用于部分与运动神经元相关的秀丽隐杆线虫运动缺陷分类任务。