scDECO: A novel statistical framework to identify differential co-expression gene combinations systematically using single-cell RNA sequencing data

项目摘要 最近的单细胞RNA测序(ScRNAseq)研究揭示了复杂的肿瘤 生态系统的特点是不同类型的细胞和 不同的转录程序。差异共表达(DC)分析正在成为 对标准基因差异表达分析(DE)的重要补充 分析数据。DC分析可以检测不同基因对之间的相关性变化 不同的调制条件。然而，大多数DC分析方法最初是 设计用于微阵列或批量RNAseq数据。有迫切的需要 开发适合以下特点的高级DC分析技术 单细胞数据、研究设计和生物学目标。 在目标1中，我们将开发一个新的、灵活的基于贝叶斯模型的框架 命名为scDECO，以提高识别DC基因组合的准确性 ScRNAseq数据。使用各种scRNAseq实验方案产生的数据，我们 将评估建议的scDECO算法并执行基准分析，以 将我们提出的方法与当前的方法进行比较。这些分析将提供一个 更好地了解这些方法的优势和局限性。 在AIM2中，我们将使用scRNAseq数据集实现scDECO算法 黑色素瘤和前列腺循环肿瘤细胞。通过识别临床上相关的集合 利用DC基因对的单细胞数据，这一发现可以促进对 肿瘤干细胞和其他细胞中转录共调控过程的研究 肿瘤微环境。此外，拟议的框架有可能 结合基因共表达改善临床疾病严重程度预测 信息纳入风险分值计算。建议的预测性能 算法将使用scRNAseq和Bulk RNAseq数据进行进一步评估。最后， 在Aim3中，将分发免费提供的R/BioConductor软件包。这个 R/生物导体环境都是生物医学研究人员非常常用的环境。 最终，这个拟议的框架将加快研究，试图理解 肿瘤中不同的协同调控转录活性。

Project Summary Recent single-cell RNA sequencing (scRNAseq) studies have revealed complex tumor ecosystems characterized by intricate interactions between heterogeneous cell types and diverse transcriptional programs. Differential co-expression (DC) analysis is emerging as a crucial complement to the standard differential expression analysis (DE) for gene profiling data. DC analysis can detect correlation changes between pairs of genes across different modulating conditions. However, most DC analysis approaches are originally designed for use on either microarray or bulk RNAseq data. There is an urgent need to develop advanced DC analytical techniques that are tailored to the characteristics of single-cell data, study design and biological objectives. In Aim 1, we will develop a novel, flexible Bayesian model-based framework named scDECO to improve the accuracy of identifying DC gene combinations using scRNAseq data. Using data generated from various scRNAseq experiment protocols, we will evaluate the proposed scDECO algorithm and perform benchmarking analyses to compare our proposed approaches to current approaches. These analyses will provide a better understanding of the advantages and limitations of these methods. In Aim2, we will implement the scDECO algorithm using scRNAseq datasets from melanoma and prostate circulating tumor cells. By identifying sets of clinically relevant DC gene pairs using single-cell data, the findings can promote understanding of the transcriptional co-regulatory processes in cancer stem-like cells and other cells in the tumor microenvironment. Furthermore, the proposed framework has the potential to improve clinical disease severity prediction by incorporating gene co-expression information into risk score calculation. The predictive performance of the proposed algorithm will be further evaluated using both scRNAseq and bulk RNAseq data. Finally, in Aim3, freely available R/Bioconductor software packages will be distributed. The R/Bioconductor environments are both very commonly used by biomedical researchers. Ultimately, this proposed framework will accelerate studies seeking to understand the differential co-regulatory transcriptional activities in tumors.