Visual Analytics for Exploration and Hypothesis Generation Using Highly MultiplexedSpatial Data of Tissues and Tumors

项目摘要 最近发展的高度多路复用亚细胞分辨率组织成像有望加速 研究肿瘤的发生、发展和免疫监视，并最终帮助发现新的 可用于临床环境的生物标志物。平行地，转录物和小分子的空间分辨测量 分子丰度达到接近单细胞分辨率。NCI项目，如人类肿瘤图谱 网络（HTAN）正在利用这些发展来创建公共数据库（“组织库”） 与癌症基因组图谱（TCGA）的范围和目标相似。制作这些数据的最大障碍是 基础和转化癌症生物学家的常规访问不在于数据收集，而在于数据 可视化和分析。现有的软件工具设计用于培养细胞实验或苏木精和 基于曙红（H&E）的数字病理学不足以用于高复杂度数据应用，新兴工具不 满足低成本和高效的数据共享或复杂的多模态机器学习的需求， 多样的数据。因此，该癌症研究信息技术（ITCR）项目将开发，强化， 并测试符合标准的软件工具，这些工具可以可视化，注释和量化 肿瘤微环境跨越105倍的长度范围（从~100 nm到1 cm）， 一套可互操作的、基于云的MINERVA工具，也可与现有的商业和开源工具很好地配合使用。 软件主要用户群体包括细胞生物学家、显微镜学家、病理学家和肿瘤学家， 成像和组织生物学以及计算生物学家和生物信息学家的专业知识， 将图像和生物学数据集成到地图集中。这些用户在我们的实验室中密切合作， 不同的需求。因此，我们的工具将支持三个阶段的研究：（i）通过直观的方法进行初始数据探索 和易于部署的基于Web的工具;（ii）通过复杂的ML支持的视觉 分析;以及（iii）数据发布和与现有知识、数据库和地图集的集成。我们 创新将包括视觉编码，ML/AI和人机界面的最新进展， 支持人在回路分析以及解释性和探索性数据可视化。 目标1将建立轻量级的方法，用于低成本的多路IF、H&E和 HTAN和类似的国际联盟收集的空间组学数据。Aim 2将开发新的方法， 探索和分析空间数据，以生成和检验假设，重点是量化 2D全载玻片和高分辨率3D图像中的形态学和细胞与细胞相互作用。目标3将扩大我们的 MINERVA平台可实现跨不同受众和数据类型的协作分析和数据共享 更好地了解肿瘤结构如何随着疾病进展和治疗而变化。我们一起， 软件将促进从复杂的多模态数据集有效地生成和检验假设， 其注释和分布符合FAIR数据原则。

PROJECT SUMMARY The recent development of highly multiplexed subcellular resolution tissue imaging promises to accelerate research into tumor initiation, progression, and immune surveillance and ultimately aid in the discovery of new biomarkers usable in a clinical setting. In parallel, spatially resolved measurement of transcript and small molecule abundances is achieving near single-cell resolution. NCI programs such as the Human Tumor Atlas Network (HTAN) are capitalizing on these developments to create public data repositories (“Tissue Atlases”) similar in scope and ambition to The Cancer Genome Atlas (TCGA). The greatest barriers to making such data routinely accessible to basic and translational cancer biologists lie not in data collection but rather data visualization and analysis. Existing software tools designed for cultured cell experiments or hematoxylin and eosin (H&E) based digital pathology are inadequate for high-plex data applications, and emerging tools do not meet the needs of either low-cost and efficient data sharing or sophisticated multi-modal machine learning from diverse data. This Informatics Technology for Cancer Research (ITCR) project will therefore develop, harden, and test standards-compliant software tools that make it possible to visualize, annotate, and quantify features of the tumor microenvironment spanning a 105-fold range in length scale (from ~100 nm to 1 cm) by building on a suite of interoperable, cloud-based MINERVA tools that also work well with existing commercial and open-source software. Key user communities include cell biologists, microscopists, pathologists and oncologists with expertise in imaging and tissue biology and computational biologists and bioinformaticians who process and integrate image and ‘omic data into atlases. These users work closely together in our laboratories but have distinct needs. Our tools will therefore support three phases of research: (i) initial data exploration via intuitive and easy-to-deploy web-based tools; (ii) hypothesis generation and testing via sophisticated ML-enabled visual analytics; and (iii) data publication and integration with existing knowledge, databases, and atlases. Our innovations will include the latest advances in visual encoding, ML/AI, and human-computer interfaces that enable human-in-the-loop analysis and explanatory and exploratory data visualization. Aim 1 will establish light-weight methods for low-cost visualization and communication of multiplex IF, H&E, and spatial omics data collected by HTAN and similar international consortia. Aim 2 will develop new ways for deeply exploring and analyzing the spatial data for hypothesis generation and testing, with a focus on quantifying morphology and cell-cell interactions in 2D whole-slide and high-resolution 3D images. Aim 3 will expand our MINERVA platform to enable collaborative analysis and data sharing across different audiences and data types to better understand how tumor architecture changes with disease progression and treatment. Together, our software will promote efficient hypothesis generation and testing from complex, multi-modal datasets as well as their annotation and distribution in accordance with FAIR data principles.