Three representative SRT ERC samples, showing histology, expression of marker genes MBP, PCP4, and SNAP25, and spatial domains

We used #Visium to generate the FIRST spatial transcriptomics dataset of the human #entorhinal cortex. By profiling gene expression across ERC layers with #BayesSpace and #spatialLIBD we uncovered gene expression patterns that could be useful in #Alzheimers research 🧠

Top row shows spots from a blue (fill) and a green (border) capture area before using visiumStitched. Once you've stitched the data together, BayesSpace and PRECAST will find up to 12 neighbors for a spot of interest by using the array/row hex grid information: 6 blue ones and 6 green ones.

Here's an improved diagram for how #visiumStitched is able to help #BayesSpace, #PRECAST, and other array-based methods (col/row from the hex grid) find more neighboring spots when you have overlapping spots from different capture areas.

You already generated the 🤩 data, might as well use it! ^_^

Spatial cluster matching rate among overlapping spots improves when using visiumStitched. It's also better for highly variable genes (HVGs) used as input to BayesSpace, and with spatially variable genes (SVGs) identified with nnSVG when using PRECAST.

One 💯 update is how @nick-eagles.bsky.social used the overlapping spots to calculate a matching rate in spatial cluster assignments

#BayesSpace with HVGs does better, whereas #PRECAST with #nnSVG SVGs does slightly better. Both improve when using #visiumStitched to increase the number of neighbors

It can handle > 2 capture areas, adjacent ones, & facilitates downstream spatially-aware clustering with methods that use the array col / row hex grid information such as #BayesSpace & #PRECAST

The proportion of matches on overlapping spots increases 📈with #visiumStitched