Question 1

Should the project use a capture-based or imaging-based platform?

Accepted Answer

Capture methods cover the whole transcriptome without prior gene selection, which suits discovery, but many operate above single-cell resolution. Imaging methods resolve individual molecules within cells and preserve subcellular position, at the cost of a predefined panel. Discovery on a capture platform followed by targeted imaging is a common project structure.

Question 2

Why does tissue handling dominate data quality?

Accepted Answer

Degraded RNA cannot be rescued computationally, so collection time, freezing method and storage matter enormously. Section thickness affects both capture efficiency and imaging focus, and permeabilisation time must be titrated per tissue since dense and loose tissues differ substantially. Record a quality metric for every block before committing to a run.

Question 3

How serious is cell segmentation error?

Accepted Answer

It is the leading source of artefact in imaging platforms. Transcripts assigned across a boundary create mixed profiles that clustering interprets as novel intermediate cell states. Membrane or boundary staining improves assignment over nuclear expansion alone. Testing whether a surprising population survives a different segmentation approach is a necessary check.

Question 4

How is composition estimated when spots contain several cells?

Accepted Answer

Deconvolution methods estimate the proportion of each cell type per spot using a single-cell reference from matched tissue. Results depend heavily on that reference, so a mismatched or incomplete one produces confident but wrong proportions. Validate a subset of predictions against imaging of marker proteins in the same or adjacent sections.

Question 5

What statistical care do spatial claims require?

Accepted Answer

Neighbouring measurements are correlated, so tests assuming independence overstate significance badly. Use null models built by permuting labels within tissue regions rather than across the whole section, since global permutation destroys architecture that was never random. Report the null model chosen, because conclusions about co-localisation depend on it.

Question 6

How many sections and donors are needed?

Accepted Answer

Adjacent sections from one block are near-replicates and mainly measure technical variation. Independent donors, and sections sampled from different regions, are what support generalisation. Because per-sample cost is high, deciding early how many donors the central comparison requires prevents an expensive dataset that cannot answer it.

Question 7

How are spatial and single-cell datasets integrated?

Accepted Answer

Single-cell data supply depth and complete transcript coverage; spatial data supply position. Mapping approaches transfer labels from the reference onto spatial coordinates or impute unmeasured genes. Treat transferred labels as predictions requiring confirmation with independent markers, not as measurements made directly on the tissue.

Biotechnology

Biotechnology

Research Frontiers in Spatial Transcriptomics Methods