A Computational IMage Analysis Platform (CIMAP) for HuBMAP

摘要：高分辨率显微镜的发展为通过基因转录物、蛋白质或代谢物的分子成像在亚微米水平上研究细胞和组织提供了前所未有的机会。基于计算机的硬件技术和人工智能/机器学习(ML)的并行发展也为以高维研究这种多组学数据提供了一种工具。一个突出的挑战涉及这些数据的融合和对所有可能领域的融合数据的透彻理解，包括在基础科学、使用模型系统的临床或临床前研究、临床诊断、预测和药物发现方面。人类生物分子图谱计划(HuBMAP)联盟是利用多种空间分子组学技术在单细胞分辨率下生成高分辨率多组学数据的途径。连接所有这些数据类型的常见成像方式是Brightfield组织学显微镜，它价格低廉，将上述多组学数据与临床决策相结合。这项HIVE Tools计划旨在开发和实施新的机器学习管道，以便使用基于蛋白质和/或RNA的空间技术数据和同时进行的Brightfield组织学来预测Brightfield组织学图像中的细胞类型和/或状态。这将使利用这些空间组学数据作为桥梁，将组织学与高含量单细胞数据集联系起来，从而创建从组织学到不同细胞类型的生物分子的单一探索空间。作为第一步，我们将使用在HuBMAP下收集的或通过这个蜂巢团队使用Codex和空间转录(ST)生成的精选数据，并开发拟议的计算管道。我们将在生成分子数据的同一切片上演示细胞类型和细胞状态到Brightfield组织学图像的映射，以及通过配准在独立相邻切片上的映射，最后在独立验证组织切片上进行映射。我们随后将探索这种方法在其他HuBMAP器官中的应用，包括淋巴结、皮肤、肝脏和肺。我们还将开发使用我们的流水线检测到的细胞类型的3D可缩放图形，目标是开发将原子与解剖学相结合的本体框架，以客观了解参考人类图谱中的可变性。我们将与其他蜂窝团队建立协同作用，将开发的管道、工具与HuBMAP网络云门户整合为一个易于使用、即插即用的最终用户插件，通过将Brightfield组织学组织图像上传到门户网站，该插件可公开访问以量化细胞计数、类型、特征和状态。我们创新的翻译科学团队模式将从生物学和工程学科招收STEM中代表性较低的少数族裔学生，为他们提供一个指导环境，并在我们的团队和合作者中提供科学机会。这一战略将培养下一代文艺复兴科学家，他们将能够继续沿着拟议的方向进行研究，将生物学、成像和工程学的知识结合到一个研究项目中。

Abstract: Advancement in high-resolution microscopy has opened unprecedented opportunities to investigate cells and tissues spatially at sub-micron level, via molecular imaging of gene transcripts, proteins or metabolomes. Parallel advances in computer-based hardware technologies and AI/ machine learning (ML) also offer a vehicle to study such multi-omics data in high dimensionality. An outstanding challenge involves a fusion of such data and thorough understanding of the fused data in all possible domains, including in basic science, clinical or pre-clinical studies using model systems, clinical diagnosis, prognostication, and drug discovery. Human Bio-Molecular Atlas Project (HuBMAP) consortium is an avenue for generating high-resolution multi-omics data at single cell resolution using a multitude of spatial molecular omics technologies. Common imaging modality that connects all these data types is brightfield histology microscopy, which is inexpensive and integrates the above-mentioned multi-omics data with clinical decision making. This HIVE Tools proposal aims to develop and implement novel machine learning pipelines to predict cell types and/or states from brightfield histology images using spatial protein- and/or RNA-based technology data with concurrent brightfield histology. This will enable using these spatial omics data as a bridge to link histology with high content single cell data sets and thus create a single exploration space from histology to biomolecules in distinct cell types. As a first step, we will employ select data collected under HuBMAP or generated via this HIVE team using CODEX as well as spatial transcriptomics (ST), and develop the proposed computational pipeline. We will demonstrate mapping of cell types and cell states to brightfield histology images on the same section from which the molecular data are generated, as well as on the independent adjacent section via registration, and finally on an independent validation tissue section. We will subsequently explore application of this approach to other HuBMAP organs including lymph node, skin, liver and lung. We will also develop 3D scalable graphics of cell types being detected using our pipeline, with a goal to develop ontological framework integrating atoms to anatomy for an objective understanding of variability in reference human atlas. We will create synergies with other HIVE teams to integrate the developed pipelines, tools with HuBMAP web-cloud portal as an easy-to-use, plug-and-play end-user plugin that is openly accessible to quantify cell counts, types, features, as well as states via uploading brightfield histology tissue images to the portal. Our innovative translational science teams’ model will recruit underrepresented minority students in STEM from biology as well as from engineering disciplines to provide them a mentorship environment and scientific opportunities within our team and that of collaborators. This strategy will develop a next generation renaissance scientist, who will be able to continue investigating along the proposed direction combining knowledge from biology, imaging, and engineering in a single research project.