Thanks, Clarence—great question! Tissue expression reflects tumor type, health, and context. With enough diverse samples across different systems, I believe the future foundation models could generalize well, though tackling challenges like batch effects will take time and effort for sure.

This is work I’m deeply passionate about and incredibly proud of. I believe that understanding spatial biology is key to unlocking new biology and drug discovery.

With Celleporter, we can simulate user-defined virtual cells in any location to uncover hidden spatial interactions within real patient tissues. Counterfactual simulations enable us to perturb environments—such as mimicking drug effects—and observe biological responses.

OCTO-vc uses spatial transcriptomics data to create “virtual cells” in real tissue. Highlights:
• Multimodal transformer trained on nearly 40M cells from 1,000+ samples
• Predicts single-cell gene expression from spatial and patient context