bulk-q数据和单细胞数据的联合分析
初一阅读题及答案比较级和最高级随着现在研究的不断深⼊，越来越多的情况需要我们对多种数据的联合分析，其中在单细胞没有出来之前，普通转录组（bulk-q）的测序结果是⾮常多的，也解决了我们很多的⽣物学问题，单细胞技术的出现，更⾼分辨率的同时，与普通转录组的联合分析也是现在分析的⼀个关注点。在⽂章《Distinct and temporary-restricted epigenetic mechanisms regulate human αβ and γδ T cell development》中就提到了单细胞和bulk数据的联合分析，在这⾥，我们来深⼊解读⼀下联合的⽅法。（当然，这篇⽂章也做了单细胞ATAC与单细胞转录组的联合分析）
我们⾸先来看⽂章中对于联合分析的结果
怎么快速学韩语>other than
Integration of our bulk RNA-q with single-cell RNA-q data from pediatric thymus samples30 confirmed that our subts reflect the continuum of human postnatal T cell development and revealed with high confidence that most thymocytes are reprented by our bulk subts。结果内容的这⾥提到了单细胞与普通转录组的联合分析，得到如下结果：
图⽚.png
图例中的解释为：Uniform manifold approximation and projection (UMAP) of the single cell RNAq datat available from with integration of our 11 bulk RNAq subts (red diamonds). DN: CD4 CD8 double negative; DP: CD4+ CD8+; P: proliferative; Q: quiescent; T(agonist): agonist lected T
影视制作学校
cells; DP (late): Positively lected DP T cells。从这个结果来看，从这个结果来看，实点代表⼀个普通转录组的数据，通过⼀定的计算⽅法映射到了单细胞数据的UMAP图上（单细胞数据也做了详细的定义划分）。
我们⾸先来看单细胞的数据来源。
voa有声双语新闻
bds单细胞数据的来源是⽂章《A cell atlas of human thymic development defines T cell repertoire formation》
这⾥我们不展开讨论，我们关注⼀下⽂章对于数据的处理，Cells with fewer than 2000 UMI counts and 500 detected genes were considered as empty droplets and removed from the datat. Cells with more than 7000 detected genes were considered as potential doublets and removed from the datat.Scanpy (version 1.3.4) python package was ud to load the cell-gene count matrix and perform downstream analysis. Doublet detection, clustering, annotation, batch alignment, trajectory analysis, cell-cell interaction, and repertoire analysis is performed using the tools in Scanpy package complemented with some custom codes.Scanpy⽤于后续分析，scanpy1.3.4版本的时候没有多样本整合的能⼒，也就是说，作者采⽤的是cellramger aggr的⽅式整合得到矩阵，然后⽤scanpy做后续分析，批次矫正采⽤线性回归的⽅式。（这个地⽅我只是猜测，并不是⼗分的确定）。
再来看bulk的数据
bulk数据是不同发育阶段的前体细胞和T细胞（磁珠分选），如下图：
图⽚.png
接下⾥就是bulk和单细胞数据的联合
bulk和单细胞数据的联合
第⼀步，Bulk thymic RNA-q samples were downsampled 50 times ranging from 100% to 2.5% quencing depth, using the downsampleMatrix from the DropletUtils library。bulk的数据处理采⽤的是“downsample”的⽅式，这是⼀个术语，这⾥就是不同的测序深度。
tolerance第⼆步，From the publicly available scRNA-q data30,pediatric samples were retained after loading publicly available scRNA-q data into R using Scanpy43 via Reticulate.单细胞数据就是⽂献来源。
儿童英文故事第三步，Genes prent in both the scRNA-q and the bulk RNA-q data were retained and both datats were
merged（merge数据）。
第四步，ed. Using Seurat each scRNA-q donor or bulk RNA-q run was procesd parately during normalization, with highly variable gene lection, data scaling and PCA generation. Batch correction of the scRNA-q data and bulk RNA-q data was done parately using the FindIntegraall that matters
tionAnchors and IntegrateData functions.样本之间先单独预处理，然后scRNA数据与bulk的数据分开整合，specifying reference samples and implementing the reciprocal PCA method这个urat⽅法⼤家可以多借鉴，值得⼀试。
第五步，联合，k.anchors was t to 15 for the bulk RNA-q data. The scRNA-q and bulk RNA-q data were integrated using the FindTransferAnchors and TransferData functions with canonical correlation analysis.关于k.anchors等参数的功能我之前介绍过，可以看我的⽂章《单细胞数据整合分析之寻找最近邻（k.anchor、k.filter、k.score、MNN）》.这个地⽅有了label转换，After integration the data were scaled, PCA was performed and a UMAP was generated.
第六步，Using the integrated data, the single cells were classified according to their bulk RNA-q counterparts, using the LogisticRegression function, specifying the liblinear solver, from the sklearn module in R via Reticulate.依据bulk的数据对单细胞进⾏数据划分（也就是定义），这⾥涉及到⼀些算法，python的模块sklearn，⼤家感兴趣多学习⼀下。
第七步，The non-T cell-lineage cells prent in the datat were ud to specify cutoffs (mean + s.d.) for each bulk RNA-q datat, allowing filtering of less probable classifications.
通过这些过程，实现了单细胞与bulk的联合分析，⽅法值得⼤家借鉴，尤其是依据bulk对单细胞数据
进⾏的定义。
请保持愤怒，让王多鱼倾家荡产，看完了，点赞呐~~~
