Glossary: one name per thing¶

The terms this site uses, each with one meaning. A term in code font is a contract value: a name the API, the package and the model cards use literally.

The two parts of a prediction¶

The route you take decides where the first part runs.

Part	What it does	Where it runs
DeepSpot-H	turns each tile of your slide into one embedding	on your machine, or on ours
DeepSpot-M	turns those embeddings into spatial gene expression	on ours, on both routes

The names on this site¶

Some things are written more than one way, depending on where you meet them.

Thing	Written for a reader	Written as a name in code	Where you meet it
The platform	Aurora	—	this site, the API, and https://auroraomics.org
Upload on the website, without code	Aurora Direct	—	https://auroraomics.org/upload
The workspace where a result is explored	Aurora Intelligence	—	https://app.auroraomics.org
Programmatic access	Aurora API	—	https://api.auroraomics.org/v1, from Python, a shell or an agent
This site	Aurora Docs	—	https://docs.auroraomics.org
The organisation behind it	Auroraomics	—	the Terms, the Privacy Policy and the licences
The Python package	the Aurora package	`auroraomics`	`pip install auroraomics`, and `import auroraomics`
The model that predicts expression	DeepSpot-M	`deepspotm`	the catalogue on Is Aurora right for your research?, and the package index
One model release	—	the card's `id`	the `model` field of a submission
The queue a model runs on	runtime	`lane`, `lanes`	a model's card, and the discovery response

Names from the wider field, and the one this site uses instead:

Term	Not called
Embedding	vector, feature map
H&E image	histopathology image, histopathological image, pathology image
Spot	pixel of the map, map pixel
Tile	crop, patch of tissue, patch
Virtual spatial transcriptomics	virtual st, predicted spatial transcriptomics

Write each model's name in full

Cut to the family name alone, either name becomes the name of a different, earlier model with its own publication and weights.

The terms¶

Virtual spatial transcriptomics: Spatially resolved gene expression predicted from an H&E image rather than measured by an assay.
H&E image: The scanned whole-slide image of a tissue section stained with haematoxylin and eosin. It is the only image input the models read.
mpp: The slide's resolution in micrometres per pixel. The package reads it from the slide, and uses it to cut every tile to the same square of tissue.
Tile: A square region cut from the slide at a known physical size, and the unit a model reads. An archive of tiles is the patches observation kind, and a model's card states the tile size it reads.
Archive: The one file the package writes your tiles into, with the record of how each was cut. It is what a patches observation carries.
Spot: One position on the slide at which expression is predicted, one per tile. A spot is one row of the result matrix, at the centre of its tile.
Prediction: What comes back for one slide: expression at every spot, as one result file. A submission asks for one.
Embedding: The row of numbers DeepSpot-H computes from one tile. Embeddings computed on your own machine are the embeddings observation kind, submitted in place of the tiles.
Transcriptome-wide: Spanning the protein-coding transcriptome. The genes a model returns are on its card and described on What you get back.
Model card: The published document that states what a model accepts, what its input must be, which genes it returns and when it retires. API reference lists them.
Observation: What a model reads: a patches archive or an embeddings file. A submission carries exactly one, as API reference sets out.
Covariate: A fact about the sample that its image does not carry: tissue, disease or bulk_rna. A submission carries any subset the model's card accepts.
Sample: What you prepared from one slide and send: the archive, or the embeddings file. On https://auroraomics.org the word names the specimen itself. Here it names the file.
Submission: The request that creates a job: POST /v1/jobs with one model, one observation and any covariates.
Upload: The resource that carries an observation's bytes to the hosted API, in parts.
Job: The record of one submission's processing: its state, its events and its result. Integration guide follows one.
Result file: One .h5ad per job, with the same layout whichever model produced it. Its gene axis is indexed by Ensembl stable gene identifier.
Gene table: The versioned table, on the service, that maps the model's gene symbols to Ensembl identifiers. Find a gene in it.
Scope: What one key is allowed to do. A key carries only the scopes it was minted with, and an operation outside them is refused. API reference lists them.
Quota: How much one key may submit over a window. Limits states every one, and a request past one is refused.
Tier: The set of counters a key is held to. Every approved account starts on the academic tier. It is not one of the website's access tiers.
API key: The credential a program presents to the hosted API. It belongs to one account, carries its scopes and tier, and expires.
Request identifier: The identifier every response carries in X-Request-Id, and every error in request_id. Quote it in a report.
The hosted API: The service reached at the API origin with an API key. Where this site says "the service", it means the hosted API.

Research terms¶

Virtual spatial transcriptomics: Predicted spatial gene expression from an H&E image, aligned to tissue coordinates.
Predicted expression: Gene expression a model predicts at a location in the tissue.
Measured expression: Gene expression observed through a molecular assay.
Cohort: A group of samples studied together to investigate a research question.