Collaborative Research: RoL: Deep-learning framework to quantify emergent phenotypes for functional gene annotation

这项工作的目标是利用机器学习的最新优势，将细菌生物膜中细胞的集体行为与其潜在的遗传网络联系起来。细胞如何自我组织成复杂的组织是现代发育生物学中最大的难题之一，也是突现行为的一个标志性例子——复杂的模式产生于简单的相互作用成分。尽管取得了巨大的进步，但即使是研究得最好的模型系统也缺乏对从分子到细胞、组织并最终完整生物体的发育现象的新兴特性的理解。土壤细菌黄粘球菌形成的生物膜是研究涌现行为的一个很好的模型系统。在饥饿状态下，黄芽孢杆菌生物膜启动一个发育程序，在此过程中细胞聚集成丘，然后分化成不同的细胞类型。许多影响黄毒杆菌发育的基因已经被确定，但研究人员缺乏系统地了解它们在协调自组织动力学中的作用的指标。该项目旨在通过基于机器学习的基因破坏对发育影响的量化，将基因和紧急行为联系起来。本项目所制订的方法可望广泛适用。通过联合会议和学员合作等密切互动，为所有参与实验室的学生提供培训机会，将进一步增强该提案的广泛影响。此外，项目拓展将包括将3d打印显微镜带入AP生物高中教室的合作努力。将基因型与涌现的多细胞表型联系起来是21世纪生物学的重大挑战之一。缺乏可靠的指标来量化遗传扰动对涌现模式的影响，严重阻碍了我们取得进展的能力，即使是相对简单的模型系统，如黄粘球菌。存在三个主要问题：(1)单个细胞的运动具有固有的随机性，它们的集体涌现模式在不同的实验重复中表现出显著的差异；(2)发育过程中出现的突发性模式具有不可预测性，对环境条件的变化极为敏感；(3)突变株的发育表型往往是微妙的，难以表征和量化。在这些问题得到解决之前，可能很难将突变的表型影响与随机性和环境敏感性的影响分开。值得注意的是，这些问题并不是M. xanthus所独有的，因此它们的解决方案有可能在许多显示紧急多细胞行为的不同生物系统中具有变革性。最近深度学习在计算机视觉中的应用进展已经证明了这些方法处理类似问题的能力。因此，开发的方法有望应用于广泛的模型系统，就像基于深度学习的图像量化方法被应用于来自各个领域的大量图像一样。这项工作是由综合有机体系统（IOS），分子细胞生物学（MCB）和生命规则（RoL）风险基金共同资助的。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The goal of this work is to leverage recent advantages in machine learning to connect the collective behavior of cells in a bacterial biofilm to their underlying genetic networks. How cells self-organize into complex tissues is one of the greatest puzzles in modern developmental biology and a hallmark example of emergent behavior - complex patterns arising from simpler interacting components. Despite tremendous progress, even the best-studied model systems lack an understanding of the emergent properties that bridge developmental phenomena from molecules to cells, tissues, and eventually complete organisms. Biofilms formed by the soil bacterium Myxococcus xanthus are a great model system to study emergent behavior. Under starvation, an M. xanthus biofilm initiates a developmental program during which cells aggregate into mounds and then differentiate into distinct cell types. Many of the genes that influence M. xanthus development have been identified, but researchers lack metrics to systematically understand their role in coordinating self-organization dynamics. This project aims to link genes and emergent behavior through machine-learning-based quantification of the developmental impact of gene disruptions. The methodology developed in this project is expected to be broadly applicable. Broader impacts of the proposal will be further enhanced by training opportunities for students for all participating laboratories, facilitated by close interactions such as joint meetings and trainee collaborations. Furthermore, project outreach will include collaborative efforts to bring 3D-printed microscopes into AP Biology high school classrooms. Connecting genotypes to emergent multicellular phenotypes is one of the grand challenges of 21st century biology. The lack of robust metrics that quantify the effects of genetic perturbations on emergent patterns significantly impedes our ability to make progress even for relatively simple model systems such as Myxococcus xanthus. Three major problems exist: (1) individual cell movements are inherently stochastic, and their collective emergent patterns display significant variations between experimental replicates; (2) emergent patterns displayed during development are unpredictable and extremely sensitivity to changes in environmental conditions; (3) developmental phenotypes of mutant strains are often subtle and difficult to characterize and quantify. Until these problems are addressed, it may be difficult to separate the phenotypic impact of mutation from the effects of stochasticity and environmental sensitivity. Notably, these problems are not unique to M. xanthus, and therefore their solution has the potential to be transformative across many different biological systems that display emergent multicellular behaviors. Recent advances in application of deep learning in computer vision have demonstrated the power of these approaches to deal with similar problems. Therefore, developed approaches are expected to apply to a wide range of model systems, just as deep-learning-based image quantification methods are being applied to a vast array of images from a variety of fields.This work is jointly funded by Integrated Organismal Systems (IOS), Molecular Cell Biology (MCB) and the Rules of Life (RoL) venture fund.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.