On technical biases in imaging spatial transcriptomics

Imaging spatial transcriptomics (iST) has revolutionized our ability to contextualize gene expression within intact tissue architectures at single-cell and subcellular resolution. However, these platforms introduce complex technical biases that span the entire measurement pipeline -- from sample preparation to data acquisition to quantification. Here, we categorize these biases into experimental (e.g., sample preservation and spatial sampling) and computational sources (e.g., z-stack registration and field-of-view stitching). Using public datasets, we demonstrate how different biases can compound and manifest, and propose a unified count modeling framework to formalize these sources of variation. Finally, we call for a more integrated understanding of iST data origins and greater knowledge sharing within and across sectors to support the prospective development of more robust experimental design and reliable downstream analysis.