Scientific practice

Interactive graphics reveal detail, but interaction does not correct a weak analysis. Use these checks before interpreting a figure.

Differential expression

• Plot adjusted p-values, not raw p-values presented as FDR.
• State both the FDR and effect-size thresholds.
• Use the MA view to inspect abundance-dependent behavior.

Microbiome composition

• Distinguish raw counts, relative abundances, and log-ratio coordinates.
• State zero handling and pseudocount choices.
• Do not apply ordinary Euclidean methods directly to percentages without justification.

Dimension reduction

• Calculate coordinates from the supplied matrix; do not invent percentages.
• Report explained variance and treat PCA as an exploratory projection.
• Do not infer discrete biological populations from visual separation alone.

Genome-wide association

• Label the multiple-testing threshold and genome build.
• Keep exact identifiers and p-values available in hover.
• Investigate population structure and linkage before biological interpretation.

Hierarchies and counts

• A sunburst or heatmap of counts is descriptive.
• Enrichment requires a background universe, statistical test, and multiple-testing correction.

Reproducibility

• Record package versions and random seeds.
• Keep data generation separate from plotting.
• Validate schemas before rendering.
• Export a static version for long-term auditability.