Maximizing the predictive power of high-throughput, microscopy-based phenotypic screens

项目摘要 我们目前拥有前所未有的能力来描述癌症中发生的遗传和途径水平的变化。然而，临床医生缺乏更有效地治疗患者亚群所需的各种药物，减少副作用，并在出现耐药性时提供二线治疗。迫切需要大幅度增加可用于对抗癌症的药物库。 自动化显微镜和计算机视觉的进步，允许在早期药物发现中广泛使用表型分析。在这份赠款中，我们解决了两个挑战。首先，表型分析的力量已经导致越来越多的大型，不同的图像数据集。在目标1中，我们将开发机器学习方法，将联合收割机不同的数据集结合起来，以获得未表征的复合函数的准确预测。第二，表型筛选可以识别多种不同途径的候选化合物，但通常只使用单一的癌细胞系。在目标2中，我们将开发策略来识别最小的细胞系集合，以最大限度地检测小分子活性。 成功执行将：通过利用现有的表型筛选数据集并提供用于选择细胞系的合理方法来增加表型分析的能力，所述细胞系最大化发现所需途径中的命中的机会。

PROJECT SUMMARY We currently have an unprecedented ability to profile the genetic- and pathway-level changes that occur in cancer. Yet, clinicians lack the diverse arsenal of drugs needed to treat subpopulations of patients more effectively, reduce side effects and offer second-line treatment when drug resistance emerges. There is a pressing need to dramatically increase the repertoire of drugs available to fight cancer. Advances in automated microscopy and computer vision, allow the widespread use of phenotypic profiling in early drug discovery. In this grant, we address two challenges. First, the power of phenotypic profiling has led to a growing number of large, disparate image datasets. In aim 1, we will develop machine-learning approaches that combine disparate datasets to obtain accurate predictions of uncharacterized compound function. Second, phenotypic screens can identify candidate compounds across multiple, diverse pathways, but often only use a single cancer cell line. In aim 2, we will develop strategies to identify minimal collections of cell lines that maximize detection of small molecule activities. Successful execution will: increase the power of phenotypic profiling by harnessing existing phenotypic screening datasets and providing a rational approach for selecting cell lines that maximize the chance of discovering hits in desired pathways.