Why Science? | NBAMSeq, Smooth Covariate Effects and RNA-seq Counts

Section 1 of 36

1. Define the independent sample

Start with the experimental unit and the question. A polished differential-expression table is only as strong as the sampling design that produced its columns.

NBAMSeq models the logarithm of mean gene counts as sums of smooth functions of covariates. Apply that point by recording what evidence would support it and what observation would weaken it.

Archive counts, sample metadata, annotations, software versions, commands, filters and random seeds so another analyst can reconstruct the route from reads to claim.

Contents · Next section

Section 2 of 36

2. Separate counts from molecules

Treat the count matrix as a chain of measurements: extraction, library construction, sequencing, alignment or quantification, annotation and summarisation all shape the number you see.

The primary study describes simultaneous estimation of smoothing parameters and coefficients in a nested iterative method. Apply that point by recording what evidence would support it and what observation would weaken it.

Count independently collected organisms, cultures, people or specimens as biological replicates. Extra reads, lanes, re-runs and repeated software fits do not create new experimental units.

Contents · Previous section · Next section

Section 3 of 36

3. Write the contrast before analysis

Write the biological comparison, reference level and covariates before opening the results. This blocks the weak habit of designing a contrast around the most colourful plot.

Variance estimation borrows information across genes using a shrinkage approach. Apply that point by recording what evidence would support it and what observation would weaken it.

Pair adjusted evidence with direction, magnitude, uncertainty, sample-level plots and a sensitivity analysis able to contradict the headline rather than merely decorate it.

Contents · Previous section · Next section

Section 4 of 36

4. Name the uncertainty

Separate biological variability, sampling noise, technical processing and model uncertainty. Each enters the evidence differently and none is cancelled by a large feature list.

A smooth term can represent curvature without forcing a particular straight-line slope. Apply that point by recording what evidence would support it and what observation would weaken it.

Declare the tested family, planned contrasts and exclusions before highlighting genes. False-discovery control belongs to the complete decision set, not just the displayed rows.

Contents · Previous section · Next section

Section 5 of 36

5. State the central assumption

Name the assumption that lets the method learn across genes or samples. Once it is visible, diagnostics and sensitivity checks become purposeful rather than ritual.

The shape is supported only where observations cover the covariate range; edge behaviour and gaps need special caution. Apply that point by recording what evidence would support it and what observation would weaken it.

Build an evidence card naming the question, measured material, statistical unit, model, contrast, result, alternatives, validation status and narrowest defensible claim.

Contents · Previous section · Next section

Section 6 of 36

6. Keep the claim bounded

Keep the conclusion at the level measured. A computational ranking can guide validation; it does not prove cellular mechanism or clinical importance.

A visually appealing curve is an association conditional on the model, not a developmental or dose mechanism. Apply that point by recording what evidence would support it and what observation would weaken it.

Use differential expression or association when that is what was estimated. Reserve cause, mechanism, diagnosis and benefit for designs and validation that can support those stronger words.

Contents · Previous section · Next section

Section 7 of 36

7. Read the official evidence in scope

Read the primary paper beside the current official package documentation. The paper explains evaluated evidence; the package page describes the software that exists now.

NBAMSeq models the logarithm of mean gene counts as sums of smooth functions of covariates. Apply that point by recording what evidence would support it and what observation would weaken it.

Archive counts, sample metadata, annotations, software versions, commands, filters and random seeds so another analyst can reconstruct the route from reads to claim.

Contents · Previous section · Next section

Section 8 of 36

8. Build a trustworthy count matrix

Use documented non-negative integer counts with unique feature identifiers. Record genome and annotation releases, counting rules and strandedness so silent mismatches are discoverable.

The primary study describes simultaneous estimation of smoothing parameters and coefficients in a nested iterative method. Apply that point by recording what evidence would support it and what observation would weaken it.

Count independently collected organisms, cultures, people or specimens as biological replicates. Extra reads, lanes, re-runs and repeated software fits do not create new experimental units.

Contents · Previous section · Next section

Section 9 of 36

9. Create a complete sample sheet

Match every matrix column to one metadata row. Include condition, batch, subject, time, sex, processing order and pairing where scientifically relevant.

Variance estimation borrows information across genes using a shrinkage approach. Apply that point by recording what evidence would support it and what observation would weaken it.

Pair adjusted evidence with direction, magnitude, uncertainty, sample-level plots and a sensitivity analysis able to contradict the headline rather than merely decorate it.

Contents · Previous section · Next section

Section 10 of 36

10. Inspect library composition

Inspect totals, detected features, zeros and dominant genes. A few abundant features can alter the comparative scale even when no code produces an error.

A smooth term can represent curvature without forcing a particular straight-line slope. Apply that point by recording what evidence would support it and what observation would weaken it.

Declare the tested family, planned contrasts and exclusions before highlighting genes. False-discovery control belongs to the complete decision set, not just the displayed rows.

Contents · Previous section · Next section

Section 11 of 36

11. Map sample relationships

Use ordination, correlations and sample-distance views as questions, not verdicts. Investigate surprising clusters against provenance before excluding any observation.

The shape is supported only where observations cover the covariate range; edge behaviour and gaps need special caution. Apply that point by recording what evidence would support it and what observation would weaken it.

Build an evidence card naming the question, measured material, statistical unit, model, contrast, result, alternatives, validation status and narrowest defensible claim.

Contents · Previous section · Next section

Section 12 of 36

12. Filter by information

Apply a declared information filter before testing and preserve the retained universe. Filtering changes estimation and the number of decisions receiving multiplicity correction.

A visually appealing curve is an association conditional on the model, not a developmental or dose mechanism. Apply that point by recording what evidence would support it and what observation would weaken it.

Use differential expression or association when that is what was estimated. Reserve cause, mechanism, diagnosis and benefit for designs and validation that can support those stronger words.

Contents · Previous section · Next section

Section 13 of 36

13. Confirm the design can answer the question

Check that the design matrix has independent information for every requested effect. Perfect condition–batch confounding has no unique statistical solution.

NBAMSeq models the logarithm of mean gene counts as sums of smooth functions of covariates. Apply that point by recording what evidence would support it and what observation would weaken it.

Archive counts, sample metadata, annotations, software versions, commands, filters and random seeds so another analyst can reconstruct the route from reads to claim.

Contents · Previous section · Next section

Section 14 of 36

14. Choose the comparison scale

Choose scaling and offsets from biology and diagnostics rather than the number of discoveries. Save the factors because they are part of the evidence.

The primary study describes simultaneous estimation of smoothing parameters and coefficients in a nested iterative method. Apply that point by recording what evidence would support it and what observation would weaken it.

Count independently collected organisms, cultures, people or specimens as biological replicates. Extra reads, lanes, re-runs and repeated software fits do not create new experimental units.

Contents · Previous section · Next section

Section 15 of 36

15. Practise with invented evidence

The table below is a fictional classroom example, visibly labelled so it cannot be mistaken for a study result.

Variance estimation borrows information across genes using a shrinkage approach. Pair adjusted evidence with direction, magnitude, uncertainty, sample-level plots and a sensitivity analysis able to contradict the headline rather than merely decorate it.

Contents · Previous section · Next section

Section 16 of 36

16. Run the method reproducibly

Record the package release, function calls, arguments, factor levels and seeds. A reproducible command history is a scientific explanation another analyst can challenge.

A smooth term can represent curvature without forcing a particular straight-line slope. Apply that point by recording what evidence would support it and what observation would weaken it.

Declare the tested family, planned contrasts and exclusions before highlighting genes. False-discovery control belongs to the complete decision set, not just the displayed rows.

Contents · Previous section · Next section

Section 17 of 36

17. Interpret the central output

Read the method-specific statistic together with its sign and units. Confirm which level is the reference before writing increase or decrease.

The shape is supported only where observations cover the covariate range; edge behaviour and gaps need special caution. Apply that point by recording what evidence would support it and what observation would weaken it.

Build an evidence card naming the question, measured material, statistical unit, model, contrast, result, alternatives, validation status and narrowest defensible claim.

Contents · Previous section · Next section

Section 18 of 36

18. Control the testing family

Control false discoveries across all genes and planned contrasts. A threshold selected after browsing the ranking is not a prespecified decision rule.

A visually appealing curve is an association conditional on the model, not a developmental or dose mechanism. Apply that point by recording what evidence would support it and what observation would weaken it.

Use differential expression or association when that is what was estimated. Reserve cause, mechanism, diagnosis and benefit for designs and validation that can support those stronger words.

Contents · Previous section · Next section

Section 19 of 36

19. Read magnitude beside uncertainty

Show fold change or another effect beside adjusted evidence and uncertainty. Statistical detectability and biological importance are not synonyms.

NBAMSeq models the logarithm of mean gene counts as sums of smooth functions of covariates. Apply that point by recording what evidence would support it and what observation would weaken it.

Archive counts, sample metadata, annotations, software versions, commands, filters and random seeds so another analyst can reconstruct the route from reads to claim.

Contents · Previous section · Next section

Section 20 of 36

20. Plot every biological sample

Plot every independent sample behind each headline feature. Means and model lines can hide outliers, inconsistent direction and sparse support.

The primary study describes simultaneous estimation of smoothing parameters and coefficients in a nested iterative method. Apply that point by recording what evidence would support it and what observation would weaken it.

Count independently collected organisms, cultures, people or specimens as biological replicates. Extra reads, lanes, re-runs and repeated software fits do not create new experimental units.

Contents · Previous section · Next section

Section 21 of 36

21. Inspect the decisive diagnostic

Choose a diagnostic that attacks the method’s defining assumption. A dashboard is useful only when each panel can change a decision.

Variance estimation borrows information across genes using a shrinkage approach. Apply that point by recording what evidence would support it and what observation would weaken it.

Pair adjusted evidence with direction, magnitude, uncertainty, sample-level plots and a sensitivity analysis able to contradict the headline rather than merely decorate it.

Contents · Previous section · Next section

Section 22 of 36

22. Challenge the central assumption

Ask what pattern would make the core assumption false, then look for it. Scientific confidence grows from surviving serious alternatives, not repeating one fit.

A smooth term can represent curvature without forcing a particular straight-line slope. Apply that point by recording what evidence would support it and what observation would weaken it.

Declare the tested family, planned contrasts and exclusions before highlighting genes. False-discovery control belongs to the complete decision set, not just the displayed rows.

Contents · Previous section · Next section

Section 23 of 36

23. Test filtering sensitivity

Repeat a small set of defensible count filters and record changes in sign, rank and adjusted evidence.

The shape is supported only where observations cover the covariate range; edge behaviour and gaps need special caution. Apply that point by recording what evidence would support it and what observation would weaken it.

Build an evidence card naming the question, measured material, statistical unit, model, contrast, result, alternatives, validation status and narrowest defensible claim.

Contents · Previous section · Next section

Section 24 of 36

24. Test normalisation sensitivity

Compare justified normalisation choices with the same samples, feature universe and contrast. Explain consequential reversals rather than hiding them.

A visually appealing curve is an association conditional on the model, not a developmental or dose mechanism. Apply that point by recording what evidence would support it and what observation would weaken it.

Use differential expression or association when that is what was estimated. Reserve cause, mechanism, diagnosis and benefit for designs and validation that can support those stronger words.

Contents · Previous section · Next section

Section 25 of 36

25. Test sample influence

Where replication permits, leave out one biological sample at a time. A discovery that disappears with one sample needs narrower language.

NBAMSeq models the logarithm of mean gene counts as sums of smooth functions of covariates. Apply that point by recording what evidence would support it and what observation would weaken it.

Archive counts, sample metadata, annotations, software versions, commands, filters and random seeds so another analyst can reconstruct the route from reads to claim.

Contents · Previous section · Next section

Section 26 of 36

26. Test method settings

Vary only settings supported by the documentation and question. Do not tune until a preferred gene crosses a threshold.

The primary study describes simultaneous estimation of smoothing parameters and coefficients in a nested iterative method. Apply that point by recording what evidence would support it and what observation would weaken it.

Count independently collected organisms, cultures, people or specimens as biological replicates. Extra reads, lanes, re-runs and repeated software fits do not create new experimental units.

Contents · Previous section · Next section

Section 27 of 36

27. Compare an adjacent method

Compare an adjacent method after aligning inputs and contrasts. Differences are clues about assumptions, not an automatic contest with one universal winner.

Variance estimation borrows information across genes using a shrinkage approach. Apply that point by recording what evidence would support it and what observation would weaken it.

Pair adjusted evidence with direction, magnitude, uncertainty, sample-level plots and a sensitivity analysis able to contradict the headline rather than merely decorate it.

Contents · Previous section · Next section

Section 28 of 36

28. Build an evidence card

Summarise provenance, design, preprocessing, model, estimate, uncertainty, diagnostics, sensitivities and validation in one compact evidence card.

A smooth term can represent curvature without forcing a particular straight-line slope. Apply that point by recording what evidence would support it and what observation would weaken it.

Declare the tested family, planned contrasts and exclusions before highlighting genes. False-discovery control belongs to the complete decision set, not just the displayed rows.

Contents · Previous section · Next section

Section 29 of 36

29. Connect to fair testing

This is the same logic as a fair school experiment: control what can be controlled, record what cannot, repeat independent units and separate observation from explanation.

The shape is supported only where observations cover the covariate range; edge behaviour and gaps need special caution. Apply that point by recording what evidence would support it and what observation would weaken it.

Build an evidence card naming the question, measured material, statistical unit, model, contrast, result, alternatives, validation status and narrowest defensible claim.

Contents · Previous section · Next section

Section 30 of 36

30. Try a classroom investigation

A safe classroom model can make the invisible assumption tangible without pretending to reproduce a laboratory sequencing experiment.

Measure a response across a continuous gradient, fit a straight line and a hand-smoothed curve, then discuss which differences are supported by observations. The activity is a learning model, not an RNA-seq experiment or health test.

A visually appealing curve is an association conditional on the model, not a developmental or dose mechanism. Use differential expression or association when that is what was estimated. Reserve cause, mechanism, diagnosis and benefit for designs and validation that can support those stronger words.

Contents · Previous section · Next section

Section 31 of 36

31. Connect to mathematics

Ratios, logarithms, distributions, smoothing, shrinkage and multiple-testing correction turn a biological question into quantities that can be checked.

NBAMSeq models the logarithm of mean gene counts as sums of smooth functions of covariates. Apply that point by recording what evidence would support it and what observation would weaken it.

Archive counts, sample metadata, annotations, software versions, commands, filters and random seeds so another analyst can reconstruct the route from reads to claim.

Contents · Previous section · Next section

Section 32 of 36

32. Connect to computing literacy

Scripts make identifiers, filters, factor levels and versions inspectable. Computing literacy means understanding how code changes the claim, not merely making code run.

The primary study describes simultaneous estimation of smoothing parameters and coefficients in a nested iterative method. Apply that point by recording what evidence would support it and what observation would weaken it.

Count independently collected organisms, cultures, people or specimens as biological replicates. Extra reads, lanes, re-runs and repeated software fits do not create new experimental units.

Contents · Previous section · Next section

Section 33 of 36

33. Connect to education pathways

RNA-seq connects biology, statistics and computing. It can illuminate study routes in life science, biotechnology and data analysis without promising admission or a career outcome.

Variance estimation borrows information across genes using a shrinkage approach. Apply that point by recording what evidence would support it and what observation would weaken it.

Pair adjusted evidence with direction, magnitude, uncertainty, sample-level plots and a sensitivity analysis able to contradict the headline rather than merely decorate it.

Contents · Previous section · Next section

Section 34 of 36

34. Ask the family question

A useful family question is simple: how many independent biological samples support this result, and which assumption most affects the ranking?

A smooth term can represent curvature without forcing a particular straight-line slope. Apply that point by recording what evidence would support it and what observation would weaken it.

Declare the tested family, planned contrasts and exclusions before highlighting genes. False-discovery control belongs to the complete decision set, not just the displayed rows.

Contents · Previous section · Next section

Section 35 of 36

35. Use precise scientific language

Write that the analysis supports differential expression or an association under the stated design and method. Avoid cause, cure, diagnosis or mechanism unless separate evidence justifies them.

The shape is supported only where observations cover the covariate range; edge behaviour and gaps need special caution. Apply that point by recording what evidence would support it and what observation would weaken it.

Build an evidence card naming the question, measured material, statistical unit, model, contrast, result, alternatives, validation status and narrowest defensible claim.

Contents · Previous section · Next section

Section 36 of 36

36. Finish with a reproducible bundle

Deliver counts or stable accessions, metadata, annotations, code, settings, complete results, plots, sensitivity analyses, package versions and the narrowest defensible interpretation.

A visually appealing curve is an association conditional on the model, not a developmental or dose mechanism. Apply that point by recording what evidence would support it and what observation would weaken it.

Use differential expression or association when that is what was estimated. Reserve cause, mechanism, diagnosis and benefit for designs and validation that can support those stronger words.

## Source date and scope note Primary studies and current official software documentation were checked for this article on 11 October 2026. Package interfaces and dependencies can change; consult the linked current official page and guide before reproduction.

## Did You Know? A method can be valuable because it makes disagreement visible. Robust normalisation, shrinkage, smooth effects and algorithm ensembles each reveal a different way that reasonable analytical choices can change an RNA-seq result.

## Final reader checklist Confirm the biological sample count, condition or covariate definition, measured material, annotation, preprocessing, statistical unit, exact contrast, effect size, uncertainty, controls, sensitivity checks, independent validation and the boundary between association and causation.

Contents · Previous section · Continue to the Science Learning Hub