AI-Aided Tool for Day Zero Selection of High Performing Cells for Biopharma Cell Line Development

总结 随着蛋白质治疗候选物数量的增加，鉴定和分离单细胞来源的 克隆是单克隆抗体药物开发中常规且频繁进行的关键步骤 和制造。细胞系开发（CLD）中的单细胞技术经历了几个阶段：首先， 通过有限稀释将单细胞置于威尔斯孔中，然后使用FACS，最近，将高增殖 在时间推移成像和机器人工具的辅助下，将单细胞分级到微量滴定板的威尔斯孔中。但没有 迄今为止的系统可以通过细胞增殖率和药物浓度来识别和分离那些“高性能”细胞， 第0天的蛋白质生产率。 我们建议开发一种创新的工具，可以预测遗传后立即细胞生长特征， 基于高通量2D/3D细胞图像和人工智能（AI）的修改。该系统的好处 包括：1）将克隆选择时间从6周缩短到2-3天，2）增加有价值的克隆数量， 50次分析的克隆（从200到10，000）。这些好处将为制药公司节省数百美元 数百万美元，并可能挽救成千上万的生命，在蛋白质为基础的疫苗生产的情况。 我们提出的工具具有几个独特的功能，包括（i）三维成像流式细胞仪（3D-IFC） 获取3D散射和2D透射图像（以及多达6种荧光颜色的3D图像）， 每个单个细胞，（ii）细胞放置模块，其放置离开3D IFC的细胞用于随后的生长，或 遗传分析，和（iii）卷积神经网络立即对单个细胞进行分类（第零天） 分为高性能和平均性能细胞，健康和患病细胞，不同表型的细胞， 正常细胞和癌细胞，以及不同的细胞类型。有了这些能力，我们提出的系统有希望 在第0天以前所未有的速度和吞吐量为CLD识别高性能细胞。 所提出的工具和技术包含以下创新特征：（a）记录2D和3D细胞 在< 20分钟内产生超过100 K的高信息量单细胞图像，（B） 将离开成像系统的每个单细胞沉积到细胞放置平台（CPP）上， 在固体培养基板上的微毛细管阵列，以将每个细胞保持在友好和索引的环境中，（c） 使用生物信息学工具检测任何细胞缺失和错位错误，以确保映射的高准确性 细胞图像到细胞位置，以及（d）使用来自2D和3D两者的融合卷积神经网络（f-CNN 标记的和/或无标记的图像来对细胞进行分类。除了CLD，所提出的工具可以有利于药物发现， 个性化医疗和基础生物医学研究，如细胞类型/细胞图谱发现和空间 生物学

SUMMARY With the increasing number of protein therapeutic candidates, identifying and isolating single-cell derived colonies is a critical step that is conducted routinely and frequently in monoclonal antibody drug development and manufacture. Single cell technologies in cell line development (CLD) has gone through a few stages: first to place single-cells in wells by limiting dilution, then to use FACS, and more recently, to place high proliferation rate single-cells into wells of a microtiter plate, aided by time lapsed imaging and robotic tools. However, no system to date can identify and isolate those “high performance” cells, judged by cell proliferation rate and drug protein production rate at Day Zero. We propose to develop an innovative tool that can predict cell outgrowth characteristics immediately after genetic modification based on high throughput 2D/3D cell image and artificial intelligence (AI). The benefits of the system include: 1) shorten the time to clone selection from 6 weeks to 2-3 days, 2) increase the number of valuable clones analyzed by 50 times (from 200 to 10,000). These benefits will save drug companies hundreds of millions of dollars, and potentially save thousands of lives in the case of protein-based vaccine production. Our proposed tool possesses several unique capabilities, including (i) a 3D imaging flow cytometer (3D-IFC) to acquire 3D scattering and 2D transmission images (plus 3D images of up to 6 fluorescent colors) of each single cell, (ii) a cell placement module that places cells exiting the 3D IFC for subsequent outgrowth or genetic analysis, and (iii) convolutional neural network to classify individual cells immediately (Day Zero) into high-performance and average performance cells, healthy and diseased cells, cells of different phenotypes, normal and cancer cells, and different cell types. With these capabilities, our proposed system holds the promise of identifying the high performing cells at Day Zero in a unprecedent speed and throughput for CLD. The proposed tool and technique contain the following innovative features: (a) recording of 2D and 3D cell images on-the-fly to produce over 100K high information content single-cell images in < 20 minutes, (b) depositing every single cell exiting the imaging system onto a cell placement platform (CPP) consisting of a microcapillary array on a solid culture medium plate to keep each cell in a friendly and indexed environment, (c) using bioinformatic tools to detect any cell deletion and misplacement errors to assure high accuracy of mapping cell images to cell positions, and (d) using a fused convolutional neural network (f-CNN) from both 2D and 3D labelled and/or label-free images to classify cells. Besides CLD, the proposed tool can benefit drug discovery, personalized medicine, and fundamental biomedical research such as cell type/cell atlas discovery and spatial biology.