专题介绍：单细胞RNA-seq被评为2018年重大科研进展，但实际上这是老技术。2015年，商品化单细胞RNA测序流程已经建立，成果发表在Cell上。今年井喷式发文章，关注点那么高，是因为最近这项技术全面商品化了。

10k PBMCs (10x Genomics)数据

他的教程展示了使用单个样本分析典型 scRNA-seq 数据集的步骤。我们将使用 10x Genomics 支持网站提供的 PBMC 数据。该管道的 DSL1 版本中使用了相同的数据集，在 SCENIC 协议教程（此处）中进行了描述。e SCENIC protocol tutorial (here).

准备10x的输入数据

“特征/细胞矩阵（过滤）”数据是从 10x Genomics 下载的，这里。 here.

wget http://cf.10xgenomics.com/samples/cell-exp/3.0.0/pbmc_10k_v3/pbmc_10k_v3_filtered_feature_bc_matrix.tar.gz

当使用 10x 数据作为输入时，管道假定文件位于典型的 Cell Ranger 目录结构中。从 10x 网站下载处理过的计数时，情况并非如此，因此我们将它们放入正确的格式：

mkdir -p pbmc10k/outs/
tar xvf pbmc_10k_v3_filtered_feature_bc_matrix.tar.gz -C pbmc10k/outs/

结果在这里：

$ tree pbmc10k
pbmc10k
└── outs
    └── filtered_feature_bc_matrix
        ├── barcodes.tsv.gz
        ├── features.tsv.gz
        └── matrix.mtx.gz

2 directories, 3 files

因此，在下一步生成的 nextflow 配置文件中，tenx 输入通道应指向 outs 文件夹。例如：

params.data.tenx.cellranger_mex = '/home/cflerin/analysis/pbmc10k/dsl2_0.19.0/pbmc10k/outs'

Setup the VSN-pipelines project

Update the repository

拉取/更新 nextflow 缓存的 vsn-pipelines 存储库。在这里，我们使用 -r 标志来指定要使用的管道版本：

nextflow pull vib-singlecell-nf/vsn-pipelines

Build the config file

We use a combination of profiles to build the config file:

• tenx: defines the input data type
• single_sample_scenic: loads the basic parameters to run the single_sample and scenic workflows
• scenic_use_cistarget_motifs and scenic_use_cistarget_tracks: includes parameters to specify the location of the cistarget database files
• hg38: specifies the genome. Other options are: hg19, dm6, mm10.
• singularity (or docker): specifies container system to use to run the processes

nextflow config vib-singlecell-nf/vsn-pipelines \
    -profile tenx,single_sample_scenic,scenic_use_cistarget_motifs,scenic_use_cistarget_tracks,hg38,singularity \
    > pbmc10k.vsn-pipelines.complete.config

Important variables to check in the config:

• singularity.runOptions (or docker.runOptions): making sure the correct volume mounts are specified (requires the user home folder (included by default in Singularity), and the location of the data).
• params.global.project_name (optional): will control the naming of the output files.
• params.sc.scope.tree.level_${X} (optional): controls the labeling of the loom file when uploaded to the SCope viewer.
• params.sc.scanpy.filter: filtering settings for the Scanpy steps.
• params.sc.scanpy.feature_selection: controls how highly variable genes are selected.
• params.sc.scanpy.clustering: controls cluster settings. In the example here, we select two clustering resolutions by using resolutions = [0.4,0.8].

Specifying compute resource usage in the config:

• The global executor (process.executor) is set to local by default. It can be changed to qsub, etc. to run specific processes as jobs. The executor parameter can be added to specific labels to run only these processes as jobs. Typically the GRN step should be submitted as a job (compute_resources__scenic_grn).
• The number of cpus and memory usage can be adjusted for each label.

The complete config file used here is available at: pbmc10k/pbmc10k.vsn-pipelines.complete.config.

Run the VSN-pipelines project

First pass

虽然总体目标是将“最佳实践步骤”和 SCENIC 一起运行，但我们可以先跳过运行 SCENIC，专注于使过滤和预处理步骤正确。然后，我们可以继续运行资源密集型 SCENIC 步骤。即使我们创建了一个包含 single_sample 和scenery 选项的配置文件，我们也可以先运行 single_sample 工作流：

nextflow -C pbmc10k.vsn-pipelines.complete.config \
    run vib-singlecell-nf/vsn-pipelines \
    -entry single_sample

Now, the QC reports can be inspected (see out/notebooks/intermediate/pbmc10k.SC_QC_filtering_report.html, either the original ipynb, or the converted html file). The cell and gene filters can be updated by editing the config file. For example, the relevant filters used here are:

params {
    sc {
        scanpy {
            filter = {
                cellFilterMinNGenes = 200
                cellFilterMaxNGenes = 4000
                cellFilterMaxPercentMito = 0.15
                geneFilterMinNCells = 3
            }
        }
    }
}

Re-run the pipeline as many times as needed (with resume to skip alread-completed steps) to select the proper filters:

nextflow -C pbmc10k.vsn-pipelines.complete.config \
    run vib-singlecell-nf/vsn-pipelines \
    -entry single_sample

Second pass

一旦细胞和基因过滤器看起来正常，我们就可以在启用完整 SCENIC 步骤的情况下重新启动管道。这将重新运行参数已更改的任何步骤（例如过滤和下游步骤），同时在使用 -resume 选项时跳过初始转换等：

nextflow -C pbmc10k.vsn-pipelines.complete.config \
    run vib-singlecell-nf/vsn-pipelines \
    -entry single_sample_scenic \
    -resume

Results

管道完成后（在 HPC 系统上使用 15 个进程进行 SCENIC GRN 步骤大约需要 2 小时），输出将是以下文件（显示被截断）：

$ tree out
out/
├── data
│   ├── intermediate
│   │   └── [...]
│   └── pbmc10k.PBMC10k_DSL2.single_sample.output.h5ad
├── loom
│   ├── pbmc10k.SCENIC_SCope_output.loom
│   └── pbmc10k.SCope_output.loom
├── nextflow_reports
│   ├── execution_report.html
│   ├── execution_timeline.html
│   ├── execution_trace.txt
│   └── pipeline_dag.dot
├── notebooks
│   ├── intermediate
│   ├── pbmc10k.merged_report.html
│   ├── pbmc10k.merged_report.ipynb
│   ├── pbmc10k.merged_report.louvain_0.4.html
│   ├── pbmc10k.merged_report.louvain_0.4.ipynb
│   ├── pbmc10k.merged_report.louvain_0.8.html
│   └── pbmc10k.merged_report.louvain_0.8.ipynb
└── scenic
    └── pbmc10k
        ├── arboreto_with_multiprocessing
        │   ├── pbmc10k__adj.tsv
        │   └── pbmc10k.filtered.loom
        ├── aucell
        │   ├── pbmc10k__auc_mtf.loom
        │   ├── pbmc10k__auc_trk.loom
        │   └── pbmc10k.filtered.loom
        ├── cistarget
        │   ├── pbmc10k.filtered.loom
        │   ├── pbmc10k__reg_mtf.csv
        │   └── pbmc10k__reg_trk.csv
        ├── notebooks
        │   ├── SCENIC_report.html
        │   └── SCENIC_report.ipynb
        ├── SCENIC_output.loom
        └── SCENIC_SCope_output.loom

The final SCENIC output is packaged into a loom file, which includes the results of the parallel expression analysis (based on highly variable genes). This can be found at out/loom/pbmc10k.SCENIC_SCope_output.loom, and is ready to be uploaded to a SCope session. The output loom file from this analysis can be found on the SCENIC protocol SCope session.

Also included is out/data/pbmc10k.PBMC10k_DSL2.single_sample.output.h5ad, an anndata file generated by the Scanpy section of the pipeline, including the results of the expression analysis (but not results from SCENIC).