Metadata feedback from a single-cell integration perspective: batch information, sample pooling and more

[LuckyMD]

Hi!

I wanted to share some feedback on the metadata suggestions. I have several questions about unclear aspects and some suggestions for extension.

Notes/questions:

• instrument metadata: I may be naive here, but is this needed? What info does this provide that would affect the data beyond what you can identify from the data itself (e.g., reads per sample)
• Sequencing protocol and library description seem very similar. It would be helpful to clarify the differences.
• Does study-id include metadata on which service the ID is registered with?
• Library id or experiment ID: accession in which database? Can authors come up with their own ID?
• Library-extract-id: libraries can consist of many pooled samples, and there can be many sample pools with the same samples. Could this library-extract-id also be a specific sample subset of a library pool of multiple samples? It would then be a subset of a sample, if this sample was pooled differently multiple times.
• is sc vs sn part of the assay_type or the library description?

Missing:

• IMO the most important unknown is batch structure when you download existing data: could be a specific field with open text or a list of fields that often confer batch effects and an additional open text field
• What about alignment information? Are all datasets that this refers to unaligned? Otherwise genome version and gene annotations as well as alignment software settings would be quite useful.

[dbujold]

Hi @LuckyMD , thank you for the comments. We'll address your points below.

instrument metadata: I may be naive here, but is this needed? What info does this provide that would affect the data beyond what you can identify from the data itself (e.g., reads per sample)

• We agree it is not the most important FAIR property, but it is still useful, and has been requested. It provides a place to enter things about the instrument use that don't have a place elsewhere (e.g. specific parameters for an instrument run, for instance) We're now providing an "How to fill" instrument and an example for the property in the "pre_produdct_feedback" branch, do you think it makes sense?
• Can also be used to assess whether other properties are relevant, such as library_layout (in a manual fashion).

Sequencing protocol and library description seem very similar. It would be helpful to clarify the differences.

• We have improved our descriptions and have a new "How to fill" column to help with these. We hope these improvements help clarifying.

Does study-id include metadata on which service the ID is registered with?

• This is down to the implementation of the checklist.

Library id or experiment ID: accession in which database? Can authors come up with their own ID?

• This is context-dependent as well. If an archive implements the checklist, they'll want to be the ones generating those ids for them to be unique. If a given lab uses the checklist, they'll come up with their own set of ids.

Library-extract-id: libraries can consist of many pooled samples, and there can be many sample pools with the same samples. Could this library-extract-id also be a specific sample subset of a library pool of multiple samples? It would then be a subset of a sample, if this sample was pooled differently multiple times.

• Good point about sample pools. Yes we will need to address that.

is sc vs sn part of the assay_type or the library description?

• “single cell” ought to be in the assay type, with a free text description in the library_description

[dbujold]

Todo: how to properly handle identifiers for pools.

[dbujold]

It'll be difficult to make a quick change for this and ensure it's been done well. We suggest keeping pools IDs for a future version in order to have a discussion with the group on how to do that properly. It will fit well with the assay categories discussions that will occur for v2.