A Comprehensive Benchmarking of Spatial Deconvolution andDomain Detection Methods Across Diverse Tissues and SpatialTranscriptomic Technologies

Why It Matters

Benchmarking clarifies which computational tools reliably reconstruct cellular composition and tissue architecture, accelerating research and reducing costly trial‑and‑error in spatial transcriptomics pipelines.

Key Takeaways

• Cell2location, RCTD, SONAR top deconvolution across tissues.
• PROST, BASS, SpaceFlow lead domain detection benchmarks.
• Performance hinges on tissue architecture, technology, and cell-type diversity.
• SynthST simulator creates realistic spatial cell-type distributions for testing.
• Practical guidelines help users choose optimal methods for varied experiments.

Pulse Analysis

Spatial transcriptomics has become a cornerstone for mapping gene expression at cellular resolution, yet the analytical bottleneck remains steep. Researchers must untangle mixed‑cell signals (spatial deconvolution) and delineate coherent tissue regions (domain detection). Prior to spDDB, the field lacked a systematic, cross‑technology comparison, forcing labs to rely on fragmented reports or ad‑hoc testing. This gap slowed adoption of emerging platforms such as Visium, Slide‑seq, and Stereo‑seq, and left many high‑throughput studies vulnerable to methodological bias.

The spDDB framework addresses these challenges by curating 37 diverse datasets and deploying a suite of 21 deconvolution and 18 domain‑detection algorithms. Its novel SynthST simulator, built on a deep graph‑attention autoencoder, generates synthetic spatial maps that preserve realistic cell‑type neighborhoods, enabling objective performance scoring. Alongside traditional accuracy metrics, the study introduces a bivariate Geary’s C statistic to capture spatial autocorrelation, as well as rare‑cell and cell‑shape assessments. Results consistently highlight Cell2location, RCTD, and SONAR for deconvolution, while PROST, BASS, and SpaceFlow excel in domain detection, though each method’s strength is modulated by tissue complexity and platform resolution.

For biotech firms, academic labs, and clinical consortia, these insights translate into faster, more cost‑effective project timelines. By following the provided guidelines, users can match their experimental design—whether profiling a heterogeneous tumor microenvironment or a developing brain—to the most suitable computational tool, reducing false‑positive region calls and improving downstream functional interpretation. Moreover, the open‑source spDDB repository invites community contributions, promising iterative refinement as new spatial technologies emerge, and cementing its role as a reference standard for the next generation of spatial omics research.