Applicability of scRNA-seq integration methods to spatial transcriptomics

Determine how well integration methods developed for bulk and single-cell RNA-sequencing—such as Mutual Nearest Neighbors (MNN), Harmony, canonical correlation analysis-based anchor methods, and deep generative models—perform when applied to integrate spatially resolved transcriptomics datasets, given intrinsic differences in experimental protocols and the biological context of spatial data.

Background

Integration strategies like MNN, Harmony, and CCA-based anchors were designed for scRNA-seq, where observations correspond to single cells and share consistent measurement contexts across datasets. Spatially resolved transcriptomics (SRT) differs fundamentally due to heterogeneous observational units (cells, spots, subcellular grids) and variable biological units captured within each observation, as well as modality-specific technical factors.

These differences may violate assumptions underlying scRNA-seq integration methods, potentially yielding spurious alignments or biased embeddings. Establishing whether and how these methods translate to SRT is necessary for reliable atlas-scale integration.