Inquiry About Spotiphy’s Deconvolution Pipeline for Visium HD (8µm) Data

[zhenhuapku]

Dears,

First, I would like to congratulate you on developing such an efficient and innovative method for spatial transcriptomics analysis. Your work on Spotiphy has provided invaluable tools for the community, and we are particularly impressed by its performance in handling high-resolution data like Visium HD.

I am writing to kindly inquire about a few technical details regarding the deconvolution pipeline for 8µm Visium HD data:

Single-cell Reference and Marker Genes
From my understanding, Spotiphy requires single-cell RNA-seq data as input for the initial deconvolution step. In our analysis, we noticed that the differentially expressed genes (DEGs) identified for deconvolution might lack specificity for certain cell types. Would it be possible (or even recommended) to manually supplement the DEG list with well-established marker genes from the literature? If so, could you advise on the best way to integrate these additional markers without biasing the results?

Nuclear Segmentation for 8µm Data
For Visium HD (8µm resolution), we are curious about the necessity of nuclear segmentation. Since each spot may still correspond to a single cell after segmentation, could you clarify whether this step serves other purposes (e.g., improving signal localization, resolving overlapping nuclei, or refining cell-type annotation)?

Looking forward to your reply.

Best regards,

[ZiqianZheng]

Hi zhenhuapku,

Thank you very much for your kind words about Spotiphy and for taking the time to explore its performance on Visium HD data. Sorry for the delayed response. Here are my answers to your questions.

Supplementing DEGs with literature-derived marker genes

You’re correct that the initial DEG list—while statistically meaningful—can sometimes underrepresent well‐known markers for rare or closely related cell types. It is both possible and sometimes beneficial to manually augment the computationally derived DEGs with curated markers from the literature. However, we have not yet established a definitive guideline for integrating these genes. We recommend running the deconvolution twice—once with DEGs alone and once with the combined list—and then comparing cell-type proportions in well-characterized regions. We are also exploring alternative approaches that are more robust to the choice of gene set.

Nuclear segmentation for 8 µm Visium HD

Segmentation in Spotiphy serves two main purposes: 1. To count the number of cells within each spot. 2. To determine the spatial location of each cell. Therefore, you're correct that if Visium HD is used and each spot is already small enough to approximate single-cell resolution, nuclear segmentation may not be necessary. However, in some Visium HD datasets, the number of UMIs detected at each location may be insufficient. In such cases, researchers might choose to combine multiple neighboring spots to increase signal. When this is done, segmentation becomes useful again, as the combined region could contain multiple nuclei.

Please feel free to reach out if you have any further questions. I'm happy to help.

Best regards,
Ziqian Zheng