Cancer Genomics: Integrative and Scalable Solutions in R/Bioconductor

项目摘要 Bioconductor是一个由2,000多个开源软件包组成的生态系统， 生物信息学分析各种类型的基因组数据。我们母基金的目标1，“癌症基因组学： R/Bioconductor中的集成和可扩展解决方案”（7 U24 CA 180996），开发和维护 R/Bioconductor数据结构，用于表示、下游软件开发和分析 多模态癌症数据集。我们的母基金的目标3建立了实验中心的网络资源， 癌症数据资源的管理、分发、维护、可重复性和可用性， R/Bioconductor社区该提案针对数百个主要以癌症为重点的基因组和 宏基因组数据集针对基于R/Bioconductor的用途进行了优化，并包含显著的附加值 以协调和手动管理的形式对主要来源进行管理，但对于哪些实质性领域和 目前需要生物导体特定的专业知识来转化为适用于广泛使用的AI/ML的格式 软件。首先，它通过翻译现有的生物导体机器学习库来创建组学 R/Bioconductor版本的TCGA、cBioPortal、宏基因组学和其他数据集。第二，为了评估 表示和推广任何开发的模型，它采用手动策展统一注释 每个研究队列的关键特征，包括人种/种族、作为生物学变量的性别、地理 地点和招聘时间。最后，它提供了导入和使用 TensorFlow、PyTorch和scikit-learn中的这些数据集。总的来说，这项提案将产生第一个大规模的， 独立于平台的AI/ML就绪数据存储库，用于存储多样化和高度策划的组学数据。彻底 对我们储存库中的研究和样本的少数地位进行注释，将有助于识别 对边缘化人群的偏见和健康差距。

Project Summary Bioconductor is an ecosystem of more than 2,000 open-source software packages for the reproducible bioinformatics analysis of various types of genomic data. Aim 1 of our parent grant, “Cancer Genomics: Integrative and Scalable Solutions in R/Bioconductor” (7U24CA180996), develops and maintains R/Bioconductor data structures for representation, downstream software development, and analysis of multimodal cancer datasets. Aim 3 of our parent grant establishes ExperimentHub web resources for the curation, distribution, maintenance, discoverability, and usability of cancer data resources for the R/Bioconductor community. This proposal targets hundreds of primarily cancer-focused genomic and metagenomic datasets that are optimized for R/Bioconductor-based usage and contain significant value-added over primary sources in the form of harmonization and manual curation, but for which substantial domain and Bioconductor-specific expertise is currently required to translate into formats suitable for widely used AI/ML softwares. First, it creates the Bioconductor Machine Learning Repository for Omics by translating existing R/Bioconductor versions of TCGA, cBioPortal, metagenomics, and other datasets. Second, in order to assess representation and generalizability of any models developed, it employs manual curation to uniformly annotate key characteristics of each study cohort including race/ethnicity, sex as a biological variable, geographical location, and recruitment period. Finally, it provides runnable documented examples of the import and use of these datasets in TensorFlow, PyTorch, and scikit-learn. In total, this proposal will produce the first large-scale, platform-independent, AI/ML-ready data repository for diverse and highly curated omics data. Thorough annotation on minority status of the studies and samples in our repository will facilitate the identification of biases and health disparities for marginalized populations.