Add multiome support (scATAC + scRNA)

[heylf]

Description of feature

Just wanted to put it down and mention that I am currently working on the implementation of cellranger-arc (modules + subworkflow). Maybe worth to discuss if this is fine to integrate into scranseq or if his should be an own pipeline as it requires a different sample sheet format and different input checks.

[grst]

Could you elaborate on how the samplesheet would need to look like? If it's just about having additional columns, I think it would be fine.

More general, we should think about which modalities (ATAC, CITE, VDJ, spatial, ...) we want to support in the future and which should be processed by the same workflow.

[heylf]

I will implement cellranger-arc in scrnaseq and then see how it goes. But yes, it would be nice to discuss the modalities.

[apeltzer]

Would still be great to discuss here before merging an entire subworkflow first 😉

[heylf]

Technically cellranger-arc needs a samplesheet (lib.cv) as an input which looks likes this:

fastqs,sample,library_type
/home/jdoe/runs/HNGEXSQXXX/outs/fastq_path,example,Gene Expression
/home/jdoe/runs/HNATACSQXX/outs/fastq_path,example,Chromatin Accessibility

Thus, lib.csv defines the folder locations for the scRNA and scATAC for the sample.

My thinking was the following:

1. Keep the definitions, subworkflows, and modules for the samplesheet as it is currently defined for scrnaseq
2. Add two new optional columns to the samplesheet: folder_GEX and folder_ATAC
3. Write a separate input_check_multiome.nf to create an input channel for scranseq.nf that has instead of [meta, [reads]] now [meta, [folders]]
4. Write a module and script to generate the lib.csv for cellranger based on the input channel from step 3.

This approach has two advantages:

1. It can be easily adapted for future data modalities (e.g., an additional methylation data) if this arrives at some point.
2. The user does not have to generate the lib.csv and stays with the definition of the samplesheet for scrnaseq

What are your thoughts about this? @grst and @apeltzer

[grst]

Personally I like to have all files explicitly listed in the samplesheet, also for consistency with other aligners.

Possible alternative:
Add an additional column sample_type to the samplesheet. The column is optional and if nothing is specified it assumes "gex":

sample,fastq_1,fastq_2,expected_cells,sample_type
pbmc8k,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L007_R1_001.fastq.gz,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L007_R2_001.fastq.gz,"10000",gex
pbmc8k,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L008_R1_001.fastq.gz,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L008_R2_001.fastq.gz,"10000",gex
pbmc8k,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L007_R1_001.fastq.gz,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L007_R2_001.fastq.gz,"10000",atac
pbmc8k,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L008_R1_001.fastq.gz,s3://nf-core-awsmegatests/scrnaseq/input_data/pbmc8k_S1_L008_R2_001.fastq.gz,"10000",atac

[heylf]

Works for me.

[apeltzer]

Tendency to go for the option proposed by @grst as it will not break existing solutions / setup 👍🏻

[heylf]

Perfect. Then I am on it!

[apeltzer]

A bit of a question remains for me how we should generally work with these multi-ome analysis types: scrnaseq (as per name suggests ;-)) is for sc-rna analysis, if we continue to add more types of analyses we might have overlaps with other nf-core pipelines (atacseq, ...) - maybe something I will put up for discussion on the general nf-core slack how to deal with these sort of things in the future 👍🏻

[heylf]

Thanks @apeltzer. Indeed, also for findability, because users might not immediately realize that you could use scrnaseq for scATAC and multiome.

[apeltzer]

Link to discussion on Slack also added here to do proper x-ref: https://nfcore.slack.com/archives/CE4K7FEHE/p1667987775811819

[apeltzer]

Please chime in there too - there is some opinions out there.

[grst]

The corresponding module has been merged:
nf-core/modules#3229

[grst]

both cellranger-arc and cellranger-multi are implemented.
Although cellranger-arc still needs tests (#290)