BIGDATA: Collaborative Research: IA: Large-Scale Multi-Parameter Analysis of Honeybee Behavior in their Natural Habitat

蜜蜂表现出高度复杂的行为，对我们的农业至关重要。由于蜜蜂丰富的社会组织，蜜蜂群体的整体表现和健康既取决于蜜蜂之间的成功分工，也取决于对环境的足够反应，这涉及复杂的行为模式和生物机制。在这些问题上仍有许多有待发现的地方，因为目前的研究受到我们在实验室外有效收集和分析个人行为的能力的限制。该项目开发的技术将使生物学家能够研究数千只蜜蜂在较长时间内的个体行为。它建立在创新算法和软件的基础上，以分析从现场蜂群收集的大数据。对这种规模的行为模式的研究将提供独特的信息，以促进对生物过程的了解，例如影响蜜蜂行为的昼夜节律，以及在动物和人类中发挥重要作用。开发的模型将有助于更好地了解涉及蜂群崩溃障碍的因素，从而指导未来对这种重要传粉者的威胁的研究。这项工作将通过一个多学科研究人员团队的紧密合作来完成，将计算机科学和数据科学的最新进展与生物学专业知识结合起来。它将提供机会，培训来自少数族裔的学生在这些领域的交叉点进行研究，并让600多名本科生、高中生和普通公众了解大数据方法如何为当前的科学和生态挑战做出贡献。该项目将开发一个高通量分析个体昆虫行为的平台，并对个体行为差异对蜜蜂群体表现的作用获得新的见解。联合视频和传感器数据采集将在大范围连续的时间内监测多个蜂群中被标记的个体，生成这种规模的第一个蜜蜂活动数据集。将开发算法和软件，以利用高性能计算设施来执行对这些海量数据集的分析。半监督机器学习将利用可用的大量数据来促进为参数创建新的检测器，如携带花粉的蜜蜂或风扇行为，目前这些参数是手动注释的。将开发预测模型和功能数据分析方法，以发现基于多参数和大时间尺度的个人行为模式。这些进展有望帮助揭示以前无法观察到的个体差异的机制。它们将使对蜜蜂昼夜节律的第一次大规模生物学研究成为可能，该研究基于个体在多种活动中的行为变化，而不是基于单一活动或平均水平的推理。进度、数据集和软件将在项目网站(sites.google.com/a/upr.edu/bigdbee)上与社区共享。

Honey bees exhibit highly complex behavior and are vital for our agriculture. Due to the rich social organization of bees, the overall performance and health of a bee colony depends both on a successful division of labor among the bees and on adequate reaction to the environment, which involves complex behavioral patterns and biological mechanisms. Much remains to be discovered on these matters as research is currently limited by our ability to effectively collect and analyze individual's behavior at large scale, out of the laboratory. The technology developed in this project will enable biologists to study the individual behavior of thousands of bees over extended periods of time. It builds on innovative algorithms and software to analyze big data collected from colonies in the field. Study of behavioral patterns at such scale will provide unique information to advance knowledge on biological processes such as circadian rhythms that influence bee behavior in addition to playing an important role in animals and humans. The models developed will help better understand factors involved in colony collapse disorder, thus guiding future research on threats to such an important pollinator. This work will be performed through the tight collaboration of a multi-disciplinary team of researchers to combine the latest advances in computer science and data science with expertise in biology. It will provide the opportunity to train students from underrepresented minority on research at the intersection of these fields and to reach more than 600 undergraduate students, high school students, and the general public about how the Big Data approach can contribute to current scientific and ecological challenges.The project will develop a platform for the high-throughput analysis of individual insect behaviors and gain new insights into the role of individual variations of behavior on bee colony performance. Joint video and sensor data acquisition will monitor marked individuals at multiple colonies over large continuous periods, generating the first datasets of bee activities of this kind on such a scale. Algorithms and software will be developed to take advantage of a High Performance Computing facility to perform the analysis of these massive datasets. Semi-supervised machine learning will leverage the large amount of data available to facilitate the creation of new detectors for parameters such as pollen carrying bees or fanning behavior, currently annotated manually. Predictive models and functional data analysis methods will be developed to find patterns in individual behavior based on multiple parameters and over large temporal scales. These advances are expected to help uncover mechanisms of individual variations previously unobservable. They will enable the first large scale biological study on the circadian rhythms of the bee based on the variations in behavior of individuals in multiple activities instead of reasoning on single activities or averages. Progress, datasets and software will be shared with the community on the project website (sites.google.com/a/upr.edu/bigdbee).