Add a new entity for the meshes that will be used to reconstruct meshes from meshes (for the meshing product)

[marwan-abdellah]

Description

We would like to have a new entity named SurfaceMesh as a generic type for any meshes that will be uploaded to reconstruct better meshes.

Technical implementation & specs

This is required for the meshing product.

Acceptance criteria

• [ ]SurfaceMesh entity in entity core.

Assignee

Will be added by default (the Product Owner)

Type

Feature

Projects

Select 'OBI 2025-' project (plus others as relevant)
Priority: P0 (higher), P1, or P2 (lower)
Time estimate: In days

[marwan-abdellah]

cc: @james-isbister @MWolfR @jdcourcol

[eleftherioszisis]

How is that related to EMCellMesh? What are the use cases that this will cover that the former cannot?

If we need something more general would it be possible to merge the two to avoid having multiple entities or table inheritance?

[marwan-abdellah]

@eleftherioszisis This is quite different from EMCellMesh. EMCellMesh represents a mesh of an EM dataset that is essentially used by the skeletonization software, such as Neuron, and in the future astrocytes and vasculature.

But this new entity is used as a general mesh, for example, a subcellular geometry, like ER, nucleus, or any other mesh that is not reconstructed from EM and is only used for mesh optimization. This will be used for the Meshing product.

cc: @jdcourcol

[eleftherioszisis]

Could we use in that case a SurfaceMesh entity for all with suitable subtype/classification/provenance columns?

[marwan-abdellah]

Indeed, We should have a SurfaceMesh as parent, and then EMCellMesh should inherit that.

[eleftherioszisis]

This is not what I suggested. Instead of using multiple-table inheritance, we could model everything as a single SurfaceMesh entity and use a few classification columns (e.g. mesh_type, source, processing_stage) to distinguish between different kinds of meshes.

A single table has several advantages:

• Simpler schema and queries (no inheritance joins).
• Better performance for common operations.
• Easier to evolve, as new mesh types only require a new classification value.
• More appropriate if all meshes share the same attributes and lifecycle.
• Fewer endpoints and code to maintain

Table inheritance is worth considering only when the subtypes have genuinely different attributes, relationships, or behavior, rather than simply representing different categories of the same entity.

[marwan-abdellah]

@eleftherioszisis What about the attributes? The EMCellMesh has many many attributes. For any other mesh type, it could contain many other attributes. Would you think that would be easy ?

[MWolfR]

I think before an implementation can begin in earnest, more info is needed:

• What are the workflows using or creating this entity?
• What are the different categories of this entity and how do they relate to the workflows?
• What properties are needed for this entity?

[marwan-abdellah]

@MWolfR
The functional workflow itself has been implemented in Ultraliser. We are trying to extend this to an operational workflow with OBI-1.
Inputs are meshes and outputs are meshes, and the attributes are only the structural ones of the EMCellMesh entity. This ticket is made to check the best possible solution before an implementation, so I think a meeting is needed to discuss this further.

[eleftherioszisis]

@marwan-abdellah can you please be more specific on which columns are shared?

entitycore/app/db/model.py

Lines 2143 to 2159 in d5cb498

	id: Mapped[uuid.UUID] = mapped_column(ForeignKey("scientific_artifact.id"), primary_key=True)
	em_dense_reconstruction_dataset_id: Mapped[uuid.UUID] = mapped_column(
	ForeignKey("em_dense_reconstruction_dataset.id"), index=True
	)
	release_version: Mapped[int]
	dense_reconstruction_cell_id: Mapped[BIGINT]
	generation_method: Mapped[EMCellMeshGenerationMethod]
	level_of_detail: Mapped[int]
	generation_parameters: Mapped[JSON_DICT | None]
	mesh_type: Mapped[EMCellMeshType]
	em_dense_reconstruction_dataset: Mapped[EMDenseReconstructionDataset] = relationship(
	foreign_keys=[em_dense_reconstruction_dataset_id],
	uselist=False,
	)
	__mapper_args__ = {"polymorphic_identity": __tablename__} # noqa: RUF012