Artificial intelligence enhanced cancer cell classification based organelle morphology and topology

摘要 乳腺癌是一种高度异质性的疾病，无论是表型和遗传。数量和 癌症蛋白质生物标志物的亚细胞定位用于对乳腺癌类型进行分类。转录组学， 多重成像或质谱细胞计数已被用于对乳腺肿瘤细胞异质性进行分类， 成功虽然基因组学和蛋白质组学在肿瘤细胞群体的鉴定方面取得了成功 参与转移性进展，确定患者肿瘤是否含有转移性 亚群仍然缺乏。最近，细胞器的形态和功能已被用作直接读出 单个癌细胞的功能表型状态。我们建议利用细胞器的空间背景， 特别是它们亚细胞位置和细胞器间关系（拓扑学），以将新的和不同的 转移性癌细胞亚群。我们开发了一个基于细胞器拓扑结构的细胞分类流水线 （OTCCP）首次量化亚细胞器的拓扑特征，定义为距离 每个细胞器对象和它在细胞内的所有邻居之间。根据RFA-CA-21-013（ 创新的信息学方法和算法癌症研究和管理），我们将适应或开发 机器学习和深度学习方法可加速和自动化基于OTCCP的细胞器- 基于拓扑学的癌细胞分类，以鉴定异质性肿瘤中的转移性细胞亚群， 具有潜在诊断和预后价值的原发性肿瘤。这种方法也将产生重大影响， 发现工具，以促进我们对癌细胞生物学在亚细胞水平上的理解。

ABSTRACT Breast cancer is a highly heterogenous disease, both phenotypically and genetically. The quantity and subcellular location of cancer protein biomarkers are used to classify breast cancer types. Transcriptomics, multiplexed imaging, or mass cytometry have been used to classify breast tumor cell heterogeneity with varying success. Although genomics and proteomics have been successful in the identification of tumor cell populations involved in metastatic progression, the ability to determine whether patient tumors contain metastatic subpopulations is still lacking. Recently, organelle morphology and function has been used as a direct readout of the functional phenotypic state of an individual cancer cell. We propose to use the spatial context of organelles, specifically their subcellular location and inter-organelle relationships (topology), to classify novel and distinct metastatic cancer cell subpopulations. We developed an Organelle Topology-based Cell Classification Pipeline (OTCCP) to quantify, for the first time, the topological features of subcellular organelles, defined as the distance between each organelle object and all its neighbors within a cell. Under RFA-CA-21-013 (Development of Innovative Informatics Methods and Algorithms for Cancer Research and Management), we will adapt or develop Machine learning and Deep Learning methodologies to accelerate and automate OTCCP-based organelle- based topology cancer cell classification to identify subpopulations of metastatic cells within heterogeneous primary tumors with potential diagnostic and prognostic value. This approach will also have major impact as a discovery tool to advance our understanding of cancer cell biology on a subcellular level.