Benchmark pitfalls expose need for expert-guided spatial clustering

TL;DR - Benchmarking spatial transcriptomics clustering tools is undermined by reproducibility, data, and evaluation limitations. Combining computational consensus with expert histology and cell-biology feedback can produce evaluations more faithful to multiscale biology and accelerate discovery.

• Spatial clustering benchmarks often lack reproducibility and suitable data.
• Existing evaluation strategies may not capture biology across spatial scales.
• Computational consensus alone is insufficient to overcome these limitations.
• Domain-expert feedback provides biological context for more reliable assessment.