0
$\begingroup$

Does single cell data typically have the following meta-data: the "coordinates" (e.g. on a tissue, adjacent tissues) saying where each cell in the sample was located relative to other cells? If not, is it possible to reconstruct this with other meta-data on the cells?

With the ultimate goal of working hands on with such location-tagged data, I am hoping for suggestions on the correct terms to search for, and even references to previous studies and/or easy to use public datasets.

For visual reference of the idea I have in mind, consider this picture:

enter image description here

I expect expression of any given gene in the skin cells in Condition 1 to be very different according to the location of the cell. For condition 2, I wouldn't expect the expression of a gene in the skin cells to be overly different, if at all. I want to see if I can formalize this idea using data mining techniques that were created for other purposes, but first I need the proper data.

edit: Here is an example of the format I eventually hope to work with.

Fake table 1: Each row represents measurements taken for a unique cell, while the last three columns have coordinates for its location. \begin{array}{r|lllllllll} \hline cell & phenotype1 & phenotype2 & \dots & gene1.expr & gene2.expr & \dots & x.loc.coord & y.loc.coord & z.loc.coord \\ \hline 1 & 0.69 & 1.34 & \dots & 1.91 & 0.21 & \dots & 1.12 & 0.05 & 1.09 \\ 2 & 0.34 & 0.92 & \dots & 1.74 & 2.03 & \dots & 0.57 & 0.46 & 0.24 \\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots \\ n & 1.97 & 1.3 & 0.96 & 0.19 & 0.66 & \dots & 0.25 & 0.02 & 1.27 \\ \hline \end{array}

Alternatively, if the (x,y,z) coordinates of Fake Table 1 are unavailable or not reconstructable, are there datasets which help us construct an adjacency list of cell pairs according to their location, such as Fake Table 2?

Fake Table 2: An adjacency table with pairs of cells which were next to each other when the measurement was taken. \begin{array}{r|cc} \hline pair & cell1 & cell2 \\ \hline 1 & 1 & 2 \\ 2 & 1 & 3 \\ 3 & 1 & 4 \\ 4 & 2 & 3 \\ 5 & 2 & 4 \\ \vdots & \vdots & \vdots \\ m-1 & n & n-2 \\ m & n & n-1 \\ \hline \end{array}

I come from a statistics background with a basic understanding of this type of data from talks I've attended, but not hands on experience. I mostly just want to explore this type of data to inform my future work. I am open to any type of scRNA-seq data with cell location "coordinates," if that's the correct thing to ask.

$\endgroup$
4
  • $\begingroup$ Hmmmm could you please explain what the end goal of your research is? You seem to be requesting RNAseq of skin cancer. Would thiis be HPV associated? $\endgroup$
    – M__
    Oct 28 '20 at 0:52
  • $\begingroup$ Hi, thank you for the question. I come from a statistics background and want to see if certain data mining techniques I have in mind are worth developing for this type of data, and overall for my exploration of this area for future work. So I am open to HPV and any other type of data with cell location "coordinates," if that's the correct thing to ask. $\endgroup$
    – gbrlrz017
    Oct 28 '20 at 1:02
  • $\begingroup$ Its just the images you have shown could be a HPV associated skin cancer. Okay I guess you are looking for classification based on morphology and the shift in expression would help assess the classification. Its just the question isn't really clear and skin cancers are known for their irregular morphology $\endgroup$
    – M__
    Oct 28 '20 at 1:11
  • $\begingroup$ I agree with you after I reread my question. I gave it an edit to be more clear. Thank you for the suggestion to look into the HPV morphology datasets! $\endgroup$
    – gbrlrz017
    Oct 28 '20 at 2:39
2
$\begingroup$

Does single cell data typically have the following meta-data: the "coordinates" (e.g. on a tissue, adjacent tissues) saying where each cell in the sample was located relative to other cells?

No. Typical scRNA-seq is just capturing random cells in a tube with no additional information. The technology you are looking for is spatial transcriptomics where you are measuring RNA levels at a particular location, but that is still not on the single-cell level.

If not, is it possible to reconstruct this with other meta-data on the cells?

There have been efforts to reconstruct spatial relationship. For example: Satija et al. 2015. There should be more recent approaches that I am not aware of.

$\endgroup$
1
  • $\begingroup$ Thank you for the clarification and for the reference! I'll read more about spatial transcriptomics. $\endgroup$
    – gbrlrz017
    Oct 28 '20 at 5:24
1
$\begingroup$

Is any of this what you want?

https://support.10xgenomics.com/spatial-gene-expression/datasets

$\endgroup$
1
  • $\begingroup$ Yes, I believe that is what I need. I was hoping this type of spatial data would have a clearer delineation of what cell each gene expression comes from, but that seems like a hard problem the technology hasn't quite solved yet based on what I have learned. $\endgroup$
    – gbrlrz017
    Oct 28 '20 at 23:51
0
$\begingroup$

A good solution is to use Human papilloma virus (HPV) PCR (see below) or particularly RNA expression of protein E5, E6 and E7 of the viral genome in skin cells. High expression of E5, E6 or E7 would be diagnostic of skin cancer because of over deregulation of latency of the virus. This is because the viral control regions (L1/L2? - I forget) get disrupted mostly by genome integration leading to E5-E7 disrupting the human cell cycle control.

Over expression would be diagnositic because alot (most? all?) skin cancer is a result of HPV. HPV types 16 and 18 are the key, but most work done in this area focuses on cervical cancer because cervical HPV infection has a high mortality rate. I suspect skin cancer has a comparatively low mortality rate (but is of considerable imporant of course).

Alot of work has been done on the tandem arrays of HPV genomic insertion into the human geneome disrupting the regulatory genes leading to over production of E5-E7 and again is considered important in malignancy. However, its the protein expression of (E5), E6 and E7 that is critical. It is not clear whether human genome integration resulting in a viral tandem arrays that disrupt viral proteins L1/L2 will automatically trigger cancer, or whether it is merely a high risk factor (i.e. tandem array human genome insertions could still remain benign).

$\endgroup$
1
  • $\begingroup$ Thank you for your insight, this seems like a promising direction. Here's hoping I can find a PCR dataset for HPV with subsamples from many neighboring tissue locations. $\endgroup$
    – gbrlrz017
    Oct 30 '20 at 0:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.