Define template for recording bags of markers

[dosumis]

Based on pattern used in

Tan, Shawn Zheng Kai, Huseyin Kir, Brian D. Aevermann, Tom Gillespie, Nomi Harris, Michael J. Hawrylycz, Nikolas L. Jorstad, et al. 2023. “Brain Data Standards - A Method for Building Data-Driven Cell-Type Ontologies.” Scientific Data 10 (1): 50. https://doi.org/10.1038/s41597-022-01886-2.

Needs slot for species context.

Also see PCL for example.

Can probably be done in ROBOT, but also easy to use DOSDP.

[github-actions[bot]]

This issue has not seen any activity in the past 6 months; it will be closed automatically in one year from now if no action is taken.

[scheuerm]

I would also like to have a template for this. But I would also like to make a distinction between a "bag of markers" and a set of markers that can serve as the necessary and sufficient axioms for defining a cell type class.

[emquardokus]

@scheuerm @dosumis Richard, I believe David's "bag of markers" is his jargon that means the same thing. I agree with you that I prefer this nomenclature to be explicitly stated as "a set of markers that can serve as the necessary and sufficient axioms for defining a cell type class."

[dosumis]

@scheuerm - agree we need to maker those we believe to be necessary and sufficient, but we also need to be careful about the use of OWL:EquivalentTo here - which is the strict logical way to assert necessary and sufficient conditions. In PCL, even with NS-Forest we ended up with incorrect automated classification of some rare cell types - IIRC this was because the NS-Forest algorithm was working on confidence of classification of single cells across a population. In the case of HubMap markers, any assertion that they are N&S is based on expert opinion rather than being derived from data. So - we may want a less formal way to assert this - one that allows use to separate out rigorous, algorithmically derived marker sets from those based on expert opinion.

[dosumis]

@scheuerm - general question. Is there some way we can use your pipelines to add classification confidence scores to marker sets coming from experts?

[scheuerm]

@dosumis yes, we just modified/modularized the NS-Forest code so that we can provide any set of markers along with a clustered cell x gene expression matrix and quantify classification performance. I've attached a comparison between NS-Forest marker performance on the performance using the markers reported in the Sikkema et al. paper for the HLCA lung atlas. The NS-Forest markers dramatically outperform the Sikkema markers for many of the clusters.

[dosumis]

Existing BDSO template:

https://github.com/obophenotype/brain_data_standards_ontologies/blob/master/src/patterns/dosdp-patterns/taxonomy_marker_set.yaml

Example input TSV:

https://github.com/obophenotype/brain_data_standards_ontologies/blob/master/src/patterns/data/default/CCN201912131_marker_set.tsv

This template already groups markers and records confidence and provenance. We need to add species and tissue context.

• TODO:
  • [ ] Pull across template to CL branch
  • [ ] Make draft modifications for review - including adding context for species and tissue.

[aleixpuigb]

https://github.com/obophenotype/cell-ontology/issues/2397 is overlapping this ticket.