Sparrow pipeline

The Sparrow pipeline by created by Pollaris et al. (2024). It was designed to analyze spatial transcriptomics data by doing over the following steps:

Image preprocessing

Based on the input image with the staining and the input data with the transcript coordinates, an image is created with transcript densities to visualize regions where no transcripts were detected. Next the image with the staining is cleaned: artefacts and halos are removed and all cells are made equally bright…

Cell segmentation

Segmentation is done by default by CellPose, but other algorithms are included in Sparrow.

Registration

Based on the transcript coordinates and the segmentation results, the transcripts are allocated to cells, generating a count matrix. Cells are annotated automatically using lists of cell type markers. This method of annotation was developed by the Saeys Lab and gave better results than existing algorithms.

Filtering genes and cells

• Cells are filtered based on transcript density, cells with low transcript density are typically low quality
• Check for each gene how many of its transcripts could not be allocated to cells: allows to detect missing cell types

Recap: overview of the Sparrow pipeline

The results of each step in the pipeline are stored in the SpatialData object.

Comparison to the pipelines used by the vendors

Sparrow generates better results than the pipelines of the vendors:

• The vendors do not provide cell type annotation
• The vendors pick up more cells but Sparrow picks up more cells with a high transcript density. The vendors pick up a lot of low quality cells and debris. Using the data generated by the vendors you loose 40% of the cell types.

Troubleshooting

Cells can be missed during the

• imaging: not bright enough, too bright, big halo…
• segmentation: too small, too big, hard regions with high cell density…
• registration: cells with no transcripts often due to necrosis…
• filtering: cells with a total count < 10 are removed and cells that are too small/big are removed

Effect of membrane stainings

MERFish (Vizgen) allows you to do multiple stainings (DAPI, polyT, membrane). All stainings are far from perfect and will miss cells. If you are interested in a cell type that is missed by the general stainings you can try a cell type specific staining. By doing limited manual annotation of these cells you can teach the segmentation algorithm to recognize them.