Collection of ideas regarding the workflow implementation.
There already exists a quality assesment pipeline, https://github.com/Teichlab/celloline, with restrictions to only run on Amazon AWS. I think that we can learn a lot from what they implemented there!
The core upstream module of the pipeline shall be https://github.com/CGATOxford/UMI-tools.
Here is a hypothetical upstream workflow:
| Step | Tool | Inputs | Output |
|---|---|---|---|
| Quality Control of FASTQs | FASTQC | R1 + R2 FASTQs | HTML Report |
| Quality Control - Detect Outliers (Gini Index) | http://www.morgridge.net/SinQC.html | R1 FASTQ | ugly report (PDFs + Tables) |
| Find Cell barcode whitelist (OPTIONAL) | umi_tools whitelist | R1 FASTQ | whitelist.txt |
| Extract CB/UMIs and filter CBs | umi_tools extract | R1 + R2 FASTQs + whitelist.txt | extracted FASTQs |
| Remove PCR duplicates | umi_tools dedup | ||
| Map reads | STAR, Kallisto, Salmon | extracted FASTQs | BAM |
| Assign reads to genes | featureCounts, Kallisto? | BAM + transcriptome GTF | BAM |
| Count unique reads per genes per cell | umi_tools count | BAM | Counts.txt |
We have to add http://multiqc.info/ report, combining interesting metrics on processing samples if that is possible. We are still lacking some trimming of the reads, if the quality is too bad (TrimGalore!, Cutadapt, Trimmomatic). What else are we missing?
We could either use published data, or simulate data e.g. using the methods described here:
There are various a GUI tools for downstream analysis: https://github.com/diazlab/scell, http://ilab.hawaii.edu:8111/?_state_id_=a9000d20618de32e&tab=info
One important R package is http://satijalab.org/seurat/. Among others it provides functionality for QC, normalization, dimension reduction (PCA, tSNE), cell clustering, find diff. expr. genes, assign cell type identites to clusters. Of all steps static visualizations are supported.
The major question I have here is: Which parts of the downstream analysis can we automate?
Quality Check of the cells
Normalization
Dimension Reduction (PCA, MDS, t-SNE, Diffusion Map, Graph Spatialization)
Differential Expression (DESeq2, EDGE_R)
Subpopulation Detection
Pseudo-Time Construction