The study of C. elegans behavior as a biophysical science

我们通过对模式生物秀丽线虫的跨学科和定量研究来学习行为物理学。生物科学已经开发出强大的分子工具来帮助我们理解生物成分(基因、蛋白质和细胞)如何形成行为背后的复杂网络。然而，对行为本身的研究已经远远落后，我们的实验室正在努力解决这一差距。通过使用新的行为表型、仪器和分析，我们努力产生可解释的、广泛适用的感觉行为模型。在这种追求中，我们研究了圆形线虫的感觉行为，这是这类工作的理想模型。它是一个相对简单的有机体(例如~300个神经元，~1000个细胞)，但它可以执行许多复杂的行为，包括搜索、逃逸和联想学习，以响应它对环境的测量。在这里，我描述一个例子来说明我们的综合和跨学科的方法。为了研究线虫的热“痛”反应，我们开发了一种新的测试方法，使用红外激光编程将精确的热加热水平应用到线虫身体的特定部位。使用这种基于激光的加热试验，我们展示了多参数表型在分析候选突变菌株文库的热“痛”行为方面的威力(Ghosh，2012)。使用多参数量化的“痛”行为，我们表明：1)线虫可以区分热刺激的精确度至少80微米时，集中在其中体，2)神经元PVD是中体反应所需的感觉神经元，以及3)一些基因编码谷氨酸受体和TRPV样通道可能参与这种伤害性的热传导(Mohammadi，2013)。然后，我们将这些高维行为数据及其动力学映射到较低维空间，以实现粗粒化。然后，利用这些数据，我们建立了一个关于热痛传导的定量统计模型，其中蠕虫的疼痛水平是通过阅读其行为反应的“肢体语言”来推断的。使用这个模型，我们客观地确定了由于基因、神经元或经验的扰动而导致的疼痛传导的变化(Leung，2016)。在相同的数据集上，我们采取了不同的方法，使用自动推理技术来生成一组关于这种疼痛行为的精确、可预测和可解释的微分方程式(Daniels，2018)。成功地发现了逃避反应背后的动力系统，说明了机器学习技术是如何帮助发现“行为方程”的。这项建议详细说明了我们通过量化、预测和可解释的模型普遍理解行为的努力，使用对行为的仔细测量和描述，同时整合了从遗传学、分子生物学和神经科学中提取的广泛的跨学科技术。

We study "the Physics of Behavior" through interdisciplinary and quantitative research of the model organism C. elegans. Biological sciences have developed powerful molecular tools to help us understand how biological components (genes, proteins, and cells) form the complex networks underlying behavior. However, the study of behavior itself has lagged far behind, and our lab is trying to address this gap. Through the use of novel behavioral phenotyping, instrumentation, and analysis, we strive to produce interpretable, widely applicable models of sensory behavior. In this pursuit we study the sensory behavior of the roundword C. elegans, an ideal model for this type of work. It is a relatively simple organism (e.g. ~300 neurons, ~1000 cells), but it can perform a number of complex behaviors including searching, escaping, and associative learning in response to measurements it makes of its environment. Here I describe an example to illustrate our integrated and interdisciplinary approach. To study the thermal "pain" response of C. elegans we developed a novel assay using an infra-red laser to programmatically apply a precise level of thermal heating to specific parts of C. elegans' body. Using this laser-based heating assay, we demonstrated the power of multi-parameter phenotyping on analysis of thermal "pain" behavior of a candidate mutant strain library (Ghosh, 2012). Using multi-parameter quantification of the "pain" behavior we showed that: 1) C. elegans can discriminate thermal stimuli to a precision of at least 80 microns when focused on its mid-body, 2) the neuron PVD is the sensory neuron that is required for the mid-body response, and 3) a number of genes encoding glutamate receptors and TRPV-like channels are likely involved in this nociceptive transduction of heat (Mohammadi, 2013). We then mapped this high-dimensional behavioral data and their dynamics onto a lower dimensional space to allow for coarse graining. Using this data, we then produced a quantitative, statistical model for thermal pain transduction, where the worm's level of pain is inferred by reading the "body language" of its behavioral response. Using this model we objectively determined changes in pain transduction due to perturbations of genes, neurons, or experience (Leung, 2016). On the same data set, we took a different approach and used automated inference techniques to generate a set of precise, predictive, and interpretable differential equations of this pain behavior (Daniels, 2018). This successful discovery of the dynamical system underlying the escape response illustrates how machine learning techniques can assist in the discovery of "equations of behavior." This proposal details our drive to understand behavior universally through quantitative, predictive, and interpretable models, using careful measurements and descriptions of behavior while integrating a wide range of interdisciplinary techniques drawn from genetics, molecular biology, and neuroscience.