Review papers by their claims

 Quick Explanation Copied

 Long Explanation

Evidence and central contribution

Yan, Harper, and Li review 34 SVG methods and organize them by biological target and statistical machinery. The key distinction is between overall SVGs—non-random spatial expression without external labels—cell-type-specific SVGs—spatial variation within a cell type—and spatial-domain-marker SVGs—higher expression in a predefined domain. Because these hypotheses differ, comparing methods by the number of genes detected can compare different scientific tasks rather than competing solutions. Primary taxonomy review

What the taxonomy clarifies

• Model choice is a biological assumption. Kernel methods target selected spatial scales or patterns; graph methods depend on neighborhood construction and edge weights; regression methods can adjust for covariates but are vulnerable to model misspecification.
• Inference is heterogeneous. The paper identifies 23 methods using frequentist tests: 11 dependence tests, 6 fixed-effect regression tests, and 6 random-effect tests. The remaining methods use Bayesian or ranking/index-based procedures, and the review correctly distinguishes rankings from calibrated significance claims.
• Technology changes interpretation. Imaging-based measurements can be near cellular or subcellular, whereas sequencing-based spots may contain multiple cells; regular grids and irregular point patterns also require different spatial representations. Treating all platforms identically is therefore an explicit risk, not a harmless simplification.

Critical assessment against benchmark evidence

The review’s warnings are supported by an independent benchmark of 31 real datasets spanning nine technologies plus nine scDesign3 simulations. That study found limited agreement in significant calls, expression-level bias in several statistics, sparsity sensitivity in neighborhood-based methods, and imperfect FDR calibration for some Giotto and Moran’s-I analyses. It also found that approximately 900–1100 selected SVGs gave the best clustering performance in one E9.5 mouse-embryo analysis, illustrating that downstream utility is not equivalent to statistical significance. Independent SVG benchmark

Blind spots and judgment

This is a strong conceptual review, but its conclusions are not themselves a systematic effect-size synthesis: the supplied paper text does not report a preregistered search strategy, formal risk-of-bias assessment, quantitative aggregation of method performance, or a reproducible scoring framework for the 34 methods. The “34” count is consequently time-dependent and may omit unpublished, newly released, or difficult-to-classify tools. The authors do, however, explicitly address several apparent weaknesses—technology and tissue differences, multi-sample analysis, double-dipping, negative controls, and benchmark design—so these should be viewed as proposed research priorities rather than overlooked limitations. Their criticism of double-dipping is especially important where domains are learned and then reused for marker testing, because feature selection and inference can be statistically dependent. Confidence: high for the taxonomy’s conceptual usefulness; moderate for universal claims about method superiority or validity.

Bottom line: use the taxonomy to define the biological question before choosing a detector; treat rankings, p-values, and downstream clustering as different evidence types; and require category-matched, platform-aware validation rather than assuming one universally best SVG method.

Author Review: Guanao YanAuthor Review: Shuo HarperAuthor Review: Jingyi Jessica Li

 BGPT Paper Review

Study Novelty

The paper does not introduce a new detector, but its explicit three-category ontology and alignment of biological targets with null hypotheses provide a useful organizing framework beyond method-by-method summaries.

Scientific Quality

The review is technically informed, distinguishes inference from ranking, identifies invalid cross-category comparisons, and discusses technology, tissue, scalability, calibration, double-dipping, and benchmarking. It is not a quantitative systematic review and does not provide a formal search or risk-of-bias protocol in the supplied text. Funding is disclosed and no competing interests are declared.

Study Generality

The framework applies across imaging- and sequencing-based SRT, tissues, spatial resolutions, and statistical paradigms, although its practical conclusions remain dependent on platform, tissue morphology, and the biological SVG definition.

Study Usefulness

The taxonomy directly improves method selection and benchmark design by separating overall, cell-type-specific, and spatial-domain-marker objectives and by emphasizing calibration, sparsity, and downstream consequences.

Study Reproducibility

The reviewed methods and conceptual classifications are described in detail, and the article reports a curated dataset list, but the review itself has no new empirical dataset, executable analysis pipeline, formal search protocol, or quantitative reproducibility assessment in the supplied text.

Explanatory Depth

The paper gives unusually clear statistical explanations of dependence, fixed-effect, and random-effect tests and connects their assumptions to SVG meaning, while offering limited quantitative resolution about comparative performance across all 34 methods.