Collaborative Research: ABI Innovation: Towards Computational Exploration of Large-Scale Neuro-Morphological Datasets

合作研究：ABI 创新：大规模神经形态数据集的计算探索

• 批准号: 2028361
• 负责人: Shuiwang Ji
• 金额: $ 14.65万
• 依托单位: Texas A&M Engineering Experiment Station
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-02-27 至 2021-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2028361&HistoricalAwards=false
• 关键词: Collaborative; Research; ABI; Innovation; Towards

Analyzing single neuron's property is a fundamental task to understand the nervous system and brain working mechanism. Investigating neuron morphology is an effective way to analyze neurons, since it plays a major role in determining neurons' properties. Recently, the ever-increasing neuron databases have greatly facilitated the research of neuron morphology. However, the sheer volume and complexity of these data pose significant challenges for computational analysis, preventing the realization of the full potential of such data. This interdisciplinary project will seek for new avenue to assemble the massive neuron morphologies and provide a unified framework for neuroscientists to explore and analyze different types of neurons. The research is able to tackle many challenges in neuroscience which are hard to solve with previous methods, including fine-grained neuron identification, latent pattern discovery and exploration, etc. The large-scale methods being developed will be particularly beneficial in the future of neuroscience, since more and more neurons are reconstructed and added to the databases. The computational methods and tools developed are very likely to be applicable for solving other bioinformatics problems, especially those dealing with large-scale datasets. The broader impact of this project not only includes educational support for undergraduate researchers and high school students, particularly women and those underrepresented groups, but also contributes to the research of neuroscience and other STEM fields.The long-term goal of this project is to develop effective computational methods and tools for neuroscientists to interactively explore large-scale neuron databases with ultra-fine-grained accuracy, in real-time. This research has a strong multidisciplinary component that involves a nexus ideas from machine learning, information retrieval, and neuroinformatics. Particularly, novel ideas will be implemented in three inter-related components through the whole framework. The first one addresses the accurate and efficient neuron reconstruction and tracing based on deep learning models. The second addresses the efficient discovery of relevant instances among large-size neuron databases via multi-modal and online binary coding methods. The third part addresses intelligent visualization and interaction schemes for knowledge discovery and mining, equipped with interactive coding that can incorporate domain experts' feedback to enhance the query algorithms for fine-tuned results. Compared with previous methods and systems, this project will open a new avenue to assist neuroscientists analyzing and exploring large-scale neuron databases with high efficiency, accuracy, and robustness. The performance of proposed methods will be validated using public neuro-morphological databases (e.g., NeuroMorpho, BigNeuron) and compared with several benchmarks. The effectiveness of the tools to be developed will be evaluated by neuroscientists on domain-specific hypothesis-driven applications. The outcome of the project will be made available at the following websites: http://webpages.uncc.edu/~szhang16/ and https://github.com/divelab/.

分析单个神经元的特性是了解神经系统和大脑工作机制的一项基本任务。研究神经元形态是分析神经元的有效方法，因为它在确定神经元的特性方面发挥着重要作用。近年来，不断增加的神经元数据库极大地促进了神经元形态学的研究。然而，这些数据的庞大数量和复杂性给计算分析带来了巨大的挑战，阻碍了这些数据的全部潜力的实现。这个跨学科项目将寻求新的途径来组装大量神经元形态，并为神经科学家探索和分析不同类型的神经元提供统一的框架。该研究能够解决神经科学中许多以前方法难以解决的挑战，包括细粒度神经元识别、潜在模式发现和探索等。正在开发的大规模方法将特别有益于神经科学的未来，因为越来越多的神经元被重建并添加到数据库中。开发的计算方法和工具很可能适用于解决其他生物信息学问题，特别是处理大规模数据集的问题。该项目更广泛的影响不仅包括为本科研究人员和高中生，特别是女性和那些代表性不足的群体提供教育支持，而且还为神经科学和其他 STEM 领域的研究做出贡献。该项目的长期目标是发展为神经科学家提供有效的计算方法和工具，以超细粒度的精度实时交互地探索大规模神经元数据库。这项研究具有强大的多学科成分，涉及机器学习、信息检索和神经信息学的联系思想。特别是，新颖的想法将在整个框架的三个相互关联的组件中实现。第一个解决了基于深度学习模型的准确高效的神经元重建和跟踪。第二个解决了通过多模式和在线二进制编码方法在大型神经元数据库中有效发现相关实例的问题。第三部分讨论了用于知识发现和挖掘的智能可视化和交互方案，配备了交互式编码，可以结合领域专家的反馈来增强查询算法以实现微调结果。与以往的方法和系统相比，该项目将为协助神经科学家高效、准确、鲁棒地分析和探索大规模神经元数据库开辟一条新途径。所提出方法的性能将使用公共神经形态学数据库（例如 NeuroMorpho、BigNeuron）进行验证，并与几个基准进行比较。神经科学家将在特定领域的假设驱动应用程序上评估要开发的工具的有效性。该项目的成果将在以下网站公布：http://webpages.uncc.edu/~szhang16/ 和 https://github.com/divelab/。

项目成果

Noise2Same: Optimizing A Self-Supervised Bound for Image Denoising

Noise2Same：优化图像去噪的自监督界限

• 发表时间: 2020-10-22
• 期刊: ArXiv
• 作者: Yaochen Xie;Zhengyang Wang;Shuiwang Ji
• 通讯作者: Shuiwang Ji

Learning Hierarchical and Shared Features for Improving 3D Neuron Reconstruction

学习分层和共享特征以改进 3D 神经元重建

• DOI: 10.1109/icdm.2019.00091
• 发表时间: 2019-11
• 期刊: Proceedings of the 19th IEEE International Conference on Data Mining
• 作者: Yuan, Hao;Zou, Na;Zhang, Shaoting;Peng, Hanchuan;Ji, Shuiwang
• 通讯作者: Ji, Shuiwang

Dense Transformer Networks for Brain Electron Microscopy Image Segmentation

用于脑电子显微镜图像分割的密集变压器网络

• DOI: 10.24963/ijcai.2019/401
• 发表时间: 2019-08-01
• 期刊: Advanced Materials Research
• 作者: Jun Li;Yongjun Chen;Lei Cai;I. Davidson;Shuiwang Ji
• 通讯作者: Shuiwang Ji

An Efficient Policy Gradient Method for Conditional Dialogue Generation

一种用于条件对话生成的高效策略梯度方法

• DOI: 10.1109/icdm.2019.00013
• 发表时间: 2019-11-01
• 期刊: 2019 IEEE International Conference on Data Mining (ICDM)
• 作者: Lei Cai;Shuiwang Ji
• 通讯作者: Shuiwang Ji

XGNN: Towards Model-Level Explanations of Graph Neural Networks

XGNN：走向图神经网络的模型级解释

• DOI: 10.1145/3394486.3403085
• 发表时间: 2020-06-03
• 期刊: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining
• 作者: Haonan Yuan;Jiliang Tang;Xia Hu;Shuiwang Ji
• 通讯作者: Shuiwang Ji