What is thy name?

A call for unison

You know that feeling when you’re frustrated over something that has absolutely no real importance, but you just can’t let it go. That’s me right now. I’m seriously bothered by the fact that every single publication seems to use a different term to denote the locations in spatial transcriptomics data.

It’s not that people aren’t using my favorite term (FYI, I don’t have one) that annoys me, but the fact that there’s no consensus on which term to use. There are beads, cells, features, voxels, pixels, spots, capture locations, and many other innovative alternatives. The current practice is confusing, and settling on one term should not be that hard.

“If names are not correct, language will not be in accordance with the truth of things.” – Confucius

My idea with this post was to present some of the different terms I’ve encountered so far (disclaimer: this is not an exhaustive list**, together with a bit of personal commentary - mainly to provide some context. To then couple this with a poll of some sort, and see what the thoughts of the community are. Perhaps this could spark some sort of discussion or reflection on the topic.

NOTE : this list was updated (2022-05-09) after posting this twitter post, where some other suggestions were presented. Thanks to all the people who contributed to a really nice discussion!

  • Bead : popularized by Slide-seq, in which beads are used to capture the transcripts within the tissue. This also resonates well with data from the HDST platform. However, it makes zero sense with data from other platforms, like Visium, where capture probes are immobilized to a solid surface, nor to imaging-based techniques like MERFISH or SeqFISH where single molecules are visualized and there’s no “capture” going on.

  • Spot : is used somewhat ambiguously in different techniques. In the ST and Visium platform, the mRNA capture locations are printed onto the surface in a fashion similar to how microarrays were constructed, which made the use of “spot” to describe the capture locations quite natural. But just as “beads” makes little sense to describe the capture locations in Visium data, “spots” makes no sense in Slide-seq data. What adds another layer of confusion is that in imaging-based methods, “spot” is sometimes used to describe the signal from a single molecule. This kind of ambiguity adds even more confusion to the already confusing situation. “Spot” is currently (2022-05-06) favored by 10x Genomics (the distributor of Visium). 10x Genomics has a large reach and, i.e., whatever term they chose, a lot of people will use it.

  • Feature : have been used by some authors to describe the spatial locations, for example, this review. Without bashing too hard on the people who prefer this term, I must confess that I find it extremely impractical to use the word “feature” to describe the “observations”. To me, a feature is the information associated with the spatial location (i.e., gene expression). Also, most analysis frameworks (e.g., Seurat) tend to call the genes (if you’re analyzing gene expression data) features. If we were to use the term feature also for our spatial locations, would we then say that we have a “feature times feature” matrix? That’s not very helpful. If confusion is something we want to avoid, we should not pursue the use of this term.

  • Pixel : a term that most of us are familiar with, but usually in the context of pictures or camera lenses. The word pixel is a portmanteau of the two words “picture” and “element” (where “pix” is used as short form of “picture”). I first saw this being used in the RCTD paper, but have also heard other people use the term in presentations. To some extent, this is a slightly more general term than “bead” or “spot”, and it works with imaging-based spatial transcriptomics methods. Also, to some extent, you could consider the Visium or Slide-seq data as an “image” with ~20,000 channels (one channel - one gene) instead of the standard three channels (RGB), but that’s a bit of a stretch. Calling the spatial data a “picture” when it’s not really image data that we’re working does not really vibe with me.

  • Voxel : is a lesser known relative to the pixel, where “vo” for volume has replaced “pix” for picture; giving us a “volume element”. My first exposure to the usage of this term to denote spatial locations was the Tangram paper. I’m not fully sure why this term was preferred over pixel, but perhaps “volume” was considered a more general term that could describe elements in both 2D and 3D methods (if you consider area as a “2D volume”). Although, I believe, for most people, volume relates very much to 3D objects; thus, potentially puzzling some people.

  • Spatial location : sometimes, often in an attempt to be method agnostic, I (and others) have used this term instead of any of the aforementioned alternatives. It definitely does the trick, it states exactly what we’re dealing with and there’s no room for confusion. But it’s terribly unsexy boring, as well as being quite long (two words!) and inconvenient to use. Sure, one could go for yet another portmanteau, but referring to spatial locations as “splocs” sounds… ehm… ridiculous.

  • Gexel : a what? Yeah you read that correct, a gexel. I stumbled upon this term when reading the 2021 MULTILAYER paper, where the authors got inspired by the pixels and conceived (I could not find any earlier use) of the “gexel”, being short for “gene expression element”. At first I kind of frowned at this term, but the more I’ve been thinking about it, the more I’ve come to like it. It’s fully method agnostic, it’s short and convenient to use, and “gene expression element” captures the essence of what our spatial locations are. The big downside to this term is its lack of generalization to other modalities, or multimodal assays where different kinds of information are registered at the same location. Perhaps one could use daxel for “data element” or spexel for “spatial expression element”, but both sound a bit “forced”.

We can summarize this is a very basic table:

Term Agnostic Descriptive Unambiguous Convenient Multimodal
bead no no yes yes yes
spot no no no yes yes
feature yes no no yes yes
pixel yes no yes yes yes
voxel yes no yes yes yes
spatial location yes yes yes no yes
gexel yes yes (+) yes yes no

Columns: Agnostic - method agnostic, Descriptive - conveys information about what it represents, Unambiguous - will not be confused with other elements of the data, Convenient - easy to use and say, Multimodal - can be used in mulitmodal assays.

P.S. I’m acutely aware of the fact that “being bothered by the inconsistency in naming of spatial locations” would be the perfect answer to “tell me you have no real problems without saying you have no real problems.” Other more important things are happening around the world, I acknowledge that 100%, but I don’t want those things to engulf my whole existence. Silly discussions must also be allowed, just as happiness must be allowed space in times of grief and sorrow.

Alma Andersson
Alma Andersson
Senior AI Scientist

My passion lies in understanding and modelling the latent structures that governs biological systems.

Related