ehennestad
commented
2 years ago I would lean towards agreeing with @lzehl that it's better to avoid the brain region when naming cell types.
The way I see it a cell type is just a way to classify cells. The key to classification of a cell has traditionally been functional or morphological, and later molecular/genetic with ever increasing number of potential cell types. But a neuron is a neuron no matter if it lives in the brain or in the periphery. And the same for subclasses of neurons.
I think the suggestion to add a small schema that produces the link between brain region and cell type is key here.
And it seems like this schema already exists. As far as I can see, cell types are only added to the tissueSample or tissueSampleCollection (and cell in the ephys module) and the information about anatomical position / region should be a part of tissueSample / cell.
Actually, I think there should be a core schema called cell, and the cell schema should at least have properties cellType and anatomicalPosition. A tissueSample would then contains cells and not cell types and thus by linkage contain information about what cell types are part of the sample and where they are located.
Also, in terms of a cell types, would it be possible to create a hierarchical schema structure, with different levels? I.e there is the very generic neuron cell type, but there is also for example interneuron, inhibitory neuron and somatostatinExpressingNeuron cell types, where all these are subclasses of each other.
The cell schema could have multiple properties related to how to classifiy it based on classification detail. Inspired by the classification of animals, there could be 'Class',' Order', 'Suborder' where 'Class' could be neuron/glia/endothelial/muscle etc, order could be inhibitory/excitatory and suborder is somatostatinExpressing / calretininExpressing etc. Or maybe it would be better to have 'FunctionalType', 'GeneExpressionType' etc as properties of the cell schema. Also from a graph database perspective this seems better.
If I go into the knowledge graph to look for interneurons, I would not find somatostatinExpressing cells etc.
Another point: If cell types should contain the region they originate from, the potential number of different cell types would explode, because where do you draw the border for the granularity of regions? And furthermore if someone recorded very specific cell types in different regions, you might end up with SomatostatinExpressingRetrosplenialCortexLayer4Cells.
Basically, as a key strategy, you do not want to combine multiple different classification levels and types into specific types because a) you end up with infinitely many combinations and b) it's harder to search at different levels of details in the knowledge graph.
tgbugs
lzehl
apdavison
There are three open issues for cell types: openMetadataInitiative/openMINDS_controlledTerms#149, openMetadataInitiative/openMINDS_controlledTerms#150 and openMetadataInitiative/openMINDS_instances#95.
We have been avoiding the discussion around how to enter cell types to the controlledTerms for too long now. @lzehl generally would like to avoid the use of brain regions in the cell type names, which I understand, but there are still issues with this:
There was a suggestion to add a small schema that produces the link between brain region and cell type, which could be interesting to follow up on now.
openMetadataInitiative/openMINDS_controlledTerms#149 and openMetadataInitiative/openMINDS_controlledTerms#150 are rather old, but still have cell types that should maybe be added, but I cannot handle openMetadataInitiative/openMINDS_instances#95 before we have landed on a final conclusion for the naming convention. We really need to land on something asap.
@tgbugs @apdavison @lzehl (feel free to link other people to this issue as well)