SIP: a Single-cell Interchangeable Pipeline

A Nextflow (DSL2) implementation of the pipeline described in Chen et al., SIP: An Interchangeable Pipeline for scRNA-seq Data Processing, bioRxiv 10.1101/456772 (2018). Tool versions are current, not the 2018 ones.

Paper section	Step	Options
2.1.0	Cells: fastq files named after cell ids, grouped by name; layout detected from the group	--cellhome
2.1.1	Read QC & preprocessing, summarised by MultiQC	--readqc fastp (default) | fastqc | afterqc
2.1.2	Strandness detection with RSeQC, used to set the downstream parameters	--strandness auto (default) | unstranded | forward | reverse
2.1.2	(Quasi-)alignment	--quant salmon (default) | star | hisat2
2.1.2–2.1.3	Gene-level counting (alignment modes); StringTie assembles, merges, re-estimates	--count featurecounts (default) | htseq | stringtie
2.1.3	Units; salmon results merged to genes with tximport	--quant_unit count | tpm (salmon), tpm | fpkm (StringTie)
2.1.4	Downstream: Seurat (clusters, t-SNE, DEGs), Monocle 2 (trajectory), scater (QC)	--using_seurat, --using_monocle, --using_scater
2.2	HPC, containers, resume, e-mail	-profile slurm, -profile singularity|docker, -resume, --email

Seurat parameters (--min_features, --max_mt, --n_pcs, --tsne_dims, --tsne_perplexity, --cluster_dims, --resolution, --de_min_pct, --heatmap_genes, --regress_vars) are listed in nextflow.config.

Input

--cellhome holds one fastq per single-end cell (<cell>.fastq.gz) or a pair per paired-end cell (<cell>_1.fastq.gz / <cell>_2.fastq.gz, _R1/_R2 also accepted), in any sub-folder layout. References: --gtf always; --transcript_fasta for salmon, --genome_fasta for STAR/HISAT2 (or a prebuilt --salmon_index / --star_index / --hisat2_index). Ensembl FASTA/GTF are expected.

Run

nextflow run main.nf -profile singularity -resume \
    --cellhome cells/ --gtf Homo_sapiens.GRCh38.115.gtf.gz \
    --transcript_fasta Homo_sapiens.GRCh38.cdna.all.fa.gz \
    --readqc fastp --quant salmon --using_seurat

Add -profile singularity,slurm to send every step to Slurm.

Output (--outdir, default sip_results/)

sip_report.pdf (summary of the run: settings, QC tables, cluster/marker tables and all figures, vector graphics), readqc/, strandness/, quant/ (salmon), assembly/ (StringTie), matrix/gene_<unit>.tsv (genes × cells), multiqc_report.html, seurat/, monocle2/, scater/.

Reproducing the paper's example (section 3)

paper/download_gse38495.sh /path/gse38495     # 8 hESC + 19 LNCaP, ~23 GB, plus Ensembl 115
nextflow run main.nf -profile singularity -params-file paper/gse38495.yaml \
    --cellhome /path/gse38495/cells --gtf /path/gse38495/ref/Homo_sapiens.GRCh38.115.gtf.gz \
    --transcript_fasta /path/gse38495/ref/Homo_sapiens.GRCh38.cdna.all.fa.gz

paper/gse38495_cells.tsv gives the cell type of each run.

reproducible/run.sh does all of this from scratch (plus the test runs below) with md5-checked inputs, digest-pinned containers and a bit-for-bit check of the results; see reproducible/README.md.

Tests

tests/make_testdata.sh builds a chr22 reference and 5 small cells (one paired-end) from the paper data; run each --readqc / --quant / --count choice on it.

Notes

• Strandness is inferred from one cell. In salmon mode its first 200k reads are mapped with salmon, and infer_experiment.py runs on that SAM against a transcript-coordinate BED; in alignment modes it runs on the first cell's BAM against the GTF.
• Monocle is pinned to 2.34: the 2.38 container ships dplyr 1.2, where Monocle's group_by_() is defunct.
• AfterQC is Python 2 and slow on full-size cells.