Breaking change: one organism per dataset with matching feature reference

[brianraymor]

Design

In 2025, the multiple species roadmap requires the following policy changes:

1. There MUST be one obs['organism_ontology_term_id'] per dataset, previously requested in Handling of multi organism datasets.
2. The var['feature_reference'] must match that obs['organism_ontology_term_id'] (with SARS-COV-2 & spike-ins also allowed)

Impact

Datasets with aggregated human and mouse

There is one aggregated dataset (Combined_human_and_mouse_limb_scRNAseq) that's Human+Mouse - https://cellxgene.cziscience.com/collections/4fefa187-5d14-4f1e-915b-c892ed320aab. The individual datasets are available in the collection.

Datasets with orthologous gene references

The following organisms appear in the corpus:

• Callithrix jacchus
• Gorilla gorilla
• Macaca mulatta
• Pan troglodytes
• Sus scrofa domesticus

allowed by the current requirement:

Organisms. Data MUST be from a Metazoan organism or SARS-COV-2 and defined in the NCBI organismal classification. For data that is neither Human, Mouse, nor SARS-COV-2, features MUST be translated into orthologous genes from the pinned Human and Mouse gene annotations.

Collection	Dataset	Organism(s)
e1fa9900-3fc9-4b57-9dce-c95724c88716	Single-nucleus transcriptome data from the dlPFC	Callithrix jacchus Homo sapiens Macaca mulatta Pan troglodytes
4dca242c-d302-4dba-a68f-4c61e7bad553	Gorilla: Great apes study	Gorilla gorilla
4dca242c-d302-4dba-a68f-4c61e7bad553	Macaque: Great apes study	Macaca mulatta
4dca242c-d302-4dba-a68f-4c61e7bad553	Marmoset: Great apes study	Callithrix jacchus
4dca242c-d302-4dba-a68f-4c61e7bad553	Chimpanzee: Great apes study	Pan troglodytes
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "medium spiny neurons" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "microglia" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "dopaminergic neurons" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "serotonergic neurons" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "basket cells" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "ependymal cells" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "cerebellar neurons" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "oligodendrocytes" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (1.5 million cell subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "glutamatergic neurons" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "vascular cells" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "astrocytes" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "GABAergic neurons" subset)	Macaca mulatta
8c4bcf0d-b4df-45c7-888c-74fb0013e9e7	A single-cell multi-omic atlas spanning the adult rhesus macaque brain (cell class "oligodendrocyte precursor cells" subset)	Macaca mulatta
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	amygdala.GAD	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	all.EPENDYMA.comp	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	hypothalamus.NEURONS	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	cortex.GAD.all	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	cortex.ASTROCYTES	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	all.MICROGLIA_MACROPHAGE.comp	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	pfc.a25.GLUT	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	hippocampus.GLUT	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	all.ASTROCYTES.comp	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	cbm.all	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	cortex.OLIGO	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	bf.NEURONS	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	striatum.SPN	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	brainstem.all	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	striatum.GAD	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	cortex.GLUT.all	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	pfc.a25.GAD	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	all.OLIGO_OPC.comp	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	hippocampus.GAD	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	amygdala.GLUT	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	thalamus.NEURONS	Callithrix jacchus
0fd39ad7-5d2d-41c2-bda0-c55bde614bdb	all.ENDOTHELIA.nocbm	Callithrix jacchus
0a77d4c0-d5d0-40f0-aa1a-5e1429bcbd7e	pig pancreatic islet cells	Sus scrofa domesticus
0a77d4c0-d5d0-40f0-aa1a-5e1429bcbd7e	Cross species map of pancreatic beta cells	Homo sapiens Mus musculus Sus scrofa domesticus
0a77d4c0-d5d0-40f0-aa1a-5e1429bcbd7e	Cross species map of pancreatic alpha cells	Homo sapiens Mus musculus Sus scrofa domesticus
6e067060-f7e4-466c-86f3-ec3dd33c0381	Inflammatory blockade in premature Rhesus macaques exposed to intra-uterine inflammation	Macaca mulatta
367d95c0-0eb0-4dae-8276-9407239421ee	Evolution of cellular diversity in primary motor cortex of human, marmoset monkey, and mouse 3-species integration inhibitory neurons	Callithrix jacchus Homo sapiens Mus musculus
367d95c0-0eb0-4dae-8276-9407239421ee	Evolution of cellular diversity in primary motor cortex of human, marmoset monkey, and mouse 3-species integration non-nuerons	Callithrix jacchus Homo sapiens Mus musculus
367d95c0-0eb0-4dae-8276-9407239421ee	Evolution of cellular diversity in primary motor cortex of human, marmoset monkey, and mouse 3-species integration excitory neurons	Callithrix jacchus Homo sapiens Mus musculus
367d95c0-0eb0-4dae-8276-9407239421ee	Evolution of cellular diversity in primary motor cortex of human, marmoset monkey, and mouse 4-species integration excitory neurons	Callithrix jacchus Homo sapiens Macaca mulatta Mus musculus

[ambrosejcarr]

From my perspective, (2) is strongly preferable. I'd be curious to understand for which fraction of datasets it would be feasible and what the estimate of the re-curation costs would be to help inform our decision.

Thinking out loud, might we make a first attempt using one of the following approaches and provide authors an opt-in opportunity to review?

Approach ideas:

1. Use an orthology service and reverse the mapping (presumably this is very straightforward, but might create errors)
2. Grab the count matrices from GEO or other archive they were submitted to, create a gene-gene correlation matrix, transfer labels where we have correlation ~= 1 (more accurate, but could fail if authors did different curation for archive submissions)

[jahilton]

It depends on what our schema says at that point (so I hesitate to dive too deep into the discuss while that's still unknown). But I think it would perfectly reasonable to say "a Dataset with only 1 organism_ontology_term_id value MUST have feature_reference to match (SARS-COV-2 & spike-ins are also allowed)". So we either get updated submission for those or they're removed.

Reversing the mapping is not a viable option in my opinion. There was already 1 mapping that obscured things and removed many genes that had no mapping. Reversing would only add another layer of obscurity (and no telling if it reverses accurately compared to the initial conversion) and you won't recover the removed genes.
I anticipate the contributors will be responsive/motivated as they will get a more accurate representation of their data.

[brianraymor]

But I think it would perfectly reasonable to say "a Dataset with only 1 organism_ontology_term_id value MUST have feature_reference to match (SARS-COV-2 & spike-ins are also allowed)".

We would also need to address the case where a dataset contains multiple organisms such as https://cellxgene.cziscience.com/collections/367d95c0-0eb0-4dae-8276-9407239421ee or https://cellxgene.cziscience.com/collections/e1fa9900-3fc9-4b57-9dce-c95724c88716.

[jahilton]

My proposal is to aim for these eventual requirements:

• If only 1 organism in obs, then the feature_reference must match (SARS-COV-2 & spike-ins are also allowed)
• If multiple organisms in obs, then the feature_reference must match 1 of them. This must be all is_primary_data:False. The obs must be represented in a single-organism Dataset.

This is similar to our philosophy with Visium - the aggregated Datasets are of value to contributors but hold no reuse so they're permitted as long as we get that data in a reusable format.

So as we get schema for each of these (allow their gene IDs and understand other standards like dev_stage) & have a timeline for the schema release, we reach out to the contributor & request their assistance in Revising their Datasets. I am hopeful for high response rate given that most find the current representation inadequate.
Once we get the non-human/-mouse Datasets listed above up to organism-specific standards, we can lock down the single-organism restriction

[jahilton]

Browsed the Datasets a little...

• All in the table above have human feature reference
• The largest dataset in 8c4bcf0d Collection has a mmulatta_homolog_ensembl_gene column in var w/ ENSMMUG.... IDs
• The lone dataset in 6e067060 Collection has a author_Rh_Gene_stable_ID column in var w/ ENSMMUG.... IDs
• The largest dataset in 0fd39ad7 Collection has a marmoset_gene_stable_id column in var w/ ENSCJAG... IDs
• The pig-only dataset in 0a77d4c0 Collection has a pig_ensembl_ID column in var w/ ENSSSCG00000027257... IDs
• The 2 x-species datasets in 0a77d4c0 Collection have mouse_ensembl_ID-mouse (ENSMUSG...) & pig_ensembl_ID-pig (ENSSSCG...) columns in var