单个细胞类型的识别是多细胞样品研究的基础。单细胞RNAseq技术允许对单个细胞进行高通量的表达分析,极大地提高了我们完成这项任务的能力。目前,大多数scna -seq数据分析都是采用无监督聚类方法进行的。根据丰富的marker基因,亚群通常被分配到不同的细胞类型。然而,这个过程是低效和武断的。在本研究中,我们提出了一个训练可扩展监督分类器的技术框架,以便在输入单细胞表达谱时就能显示单细胞的身份。通过使用多个scna -seq数据集,证明了该方法与传统方法相比具有较高的精度、鲁棒性、兼容性和可扩展性。我们使用两个模型升级的例子来演示如何实现单元类型分类器的预测演化。
一款在线做单细胞细胞类型定义的工具被开发出来了!
- 只需要输入cellranger结果即可
- 在线,操作方便
- 基于Seurat,界面其实是一个shiny项目
- Python环境
- 目前只能做人和小鼠。
操作及其简单:
需要注册:
在线网站: SuperCT
上传矩阵即可:
When uploading 10xgenomics file, you need to compress the 3 files ‘genes.tsv’, ‘barcodes.tsv’ and ‘matrix.tsv’ under ‘filtered_gene_bc_matrices’.
等待:
细胞定义结果:
导出CSV:
下面是简单介绍。
SuperCT: a supervised-learning framework for enhanced characterization of single-cell transcriptomic profiles
Overview of SuperCT framework and the high concordance between the original MCA cell-type annotations and the SuperCT predictions. (A) The workflow of SuperCT training, prediction and upgrade. (B) The two upgrades that lead to the optimized or expanded SuperCT classifier. (C) The overview of TMC data original labels in comparison with SuperCT v2m predictions.
The outperformance of SuperCT over traditional UC-based methods. (A) Discordant cell labels by TMC and by SuperCT v2m predictions in spleen tissue: monocyte/macrophage versus dendritic cell. (B) The higher signal of dendritic cell signature genes suggests the SuperCT gives more convincing labels. (C) Down-sampling makes the minor cell populations lose the power to form discernable cluster but SuperCT can still characterize the cell type. (D) The separated clusters of the same cell types derived from batch difference can still be correctly characterized by SuperCT.
This single-cell RNAseq analytical tool is to characterize the cell types of heterogeneous samples using the UMI-based single-cell RNAseq data and a supervised classifier framework. 46 types can be characterized for the more recent version. More types will be included in the future. The technical details of the model and the training strategy have been put in a write-up and the manuscript will be submitted to the bioRxiv.org in a couple of days and hopefully be published in a prestigious journal soon after. You are welcome to test this tool by submitting your own UMI matrices. The cell types can be visualized based on the layout of tSNE in the Seurat style. We also provide the visualization to view the signal of the specific cell types.
This application is under continuous development by the inclusion of more and more high-quality training data sets and high-confidence cell-type labels. Your feedback on the bugs or the drawbacks will be highly appreciated. You are also encouraged to submit your curated cell types so as to make this tool better. The further collaborative effort can be discussed in person.
Advices for the dge matrix uploading:
If you upload the dge matrix, please use gene symbol ID instead of Ensemble IDs. The human genes will be like ‘CD3D’, ‘IL10’ etc. The mouse genes will be like ‘Cd3d’, ‘Il10’ etc. A typical dge.txt file will be like the following:
dgeBefore uploading dge file, you are advised to do the following quality assurance. The duplicate gene names should be avoided. This is a major issue that causes trouble in Seurat pre-processing. In addition, you also need to double-check the total UMI count of each column (each cell) and the transcriptome complexity (how many genes are detected. >500 genes is preferred). Unfortunately, the dge matrix generated from high sequencing depth, such as Fluidigm C1 or Smarter-seq protocols may not work in here because our model is trained from cell-barcodes+UMI type of data.
Advices for 10xGenomics matrix uploading:
When uploading 10xgenomics file, you need to compress the 3 files ‘genes.tsv’, ‘barcodes.tsv’ and ‘matrix.tsv’ under ‘filtered_gene_bc_matrices’.
Frequently Asked Questions:
1: Why it takes so long to view result?
We are still working hard on the optimization of the system, especially exception processing. If you don’t get the result for more than 2 hours, it means the process is halted due to bugs. But we will frequently check the log, fix the bugs and resend the result before you know it.
2: Is my dataset secured?
Yes, we promise your data won't be misused or disclosed to any third party without your permission.
Any question or concern can be sent to the developer's email address: weilin.baylor@gmail.com
SuperCT
SuperCT: a supervised-learning framework for enhanced characterization of single-cell transcriptomic profiles
网友评论