STITCH links cellular morphology and gene expression in spatial transcriptomics

Imagine you are looking at a bustling city through a high-tech window that lets you see two things at once: the shape of every building (the cells) and the blueprints inside them (the genes). Scientists call this "spatial transcriptomics." The goal is to understand how the look of a building relates to what's happening inside it.

However, there's a problem with the tools scientists have been using. Think of it like trying to judge the style of a house by looking at a photo, but your camera is so zoomed out that it can't tell the difference between a tiny, perfectly shaped cottage and a giant, messy mansion. Current computer programs mix up the size of the cell with its shape. If a cell is just big, the computer thinks it has a different "personality" than a small one, even if they are shaped exactly the same.

Enter STITCH, a new method developed by the researchers. Here is how it works, using a simple analogy:

The "Shape-Only" Filter

Imagine you have a bag of different-sized clay animals. Some are tiny mice, some are giant elephants, but they are all molded into the exact same "mouse shape."

• Old methods would say, "The elephant is totally different from the mouse because it's huge."
• STITCH acts like a magical filter. It ignores the size completely and only looks at the outline. It uses a special mathematical trick (called TPCA) to say, "Ah, this is a mouse shape, regardless of whether it's tiny or huge." This allows scientists to see the true "personality" of the cell's shape without being fooled by its size.

What STITCH Found

The researchers tested this new "shape filter" on real biological data and found some fascinating connections:

1. Better at Finding the "Architects": When they tried to find genes that control cell shape (like an RNAi screen), STITCH was much better at spotting the right "architects" than older methods. It successfully identified the genes that, when turned off, actually changed the shape of the cell.
2. Mapping the Skin Layers: In a dataset of skin cells (keratinocytes), STITCH successfully mapped out the different layers of the skin and noticed a smooth gradient in how round or oval the cell nuclei were. A deep-learning approach (the old, size-confused method) missed these subtle patterns.
3. The "Bodyguard" Fibroblasts: In melanoma (skin cancer) samples, STITCH noticed that certain helper cells (fibroblasts) that were long and triangular tended to hang out right next to the cancer cells. These specific shapes were linked to a specific "job description" (transcriptional program) inside the cell, suggesting they were acting like bodyguards or construction workers near the danger zone.
4. The "Stretchy" Cancer Cells: In two different groups of melanoma patients, STITCH independently confirmed a known rule: cancer cells that act like "mesenchymal" cells (a type of cell that likes to move and migrate) tend to be physically larger in area. STITCH found this link on its own, proving it works across different patients and machines.

The Bottom Line

STITCH is like a new pair of glasses that lets scientists see the true shape of cells without getting distracted by their size. By doing this, it creates a clear, understandable map showing how a cell's physical appearance is linked to its genetic instructions. This helps researchers understand cell behavior across different patients and different types of tissue, all while keeping the math simple and interpretable.