Omics analysis problems

Match feature, association, and multivariate targets to defensible evidence

An analysis result needs a population meaning before it becomes a biological claim. A feature contrast, a network edge, a pseudo-value summary, and an omnibus multivariate comparison are different targets. Their uncertainty arguments must match the observation, sampling design, and estimator.

These pages build two habits: define what a feature or edge estimate means in the population, then calibrate the evidence for the claims being reported. Regression contrasts, pooling, imperfect adjustment, and residual covariance show how a plausible analysis can answer a different question from the one intended. Multivariate tests add a whole-profile question whose null and resampling unit also need to be explicit.

Multiplicity connects all of these problems. A paper may report thousands of features or millions of edges, so a collection of individually calibrated tests still needs a justified reporting rule. The multiple-testing page develops the distinction between a realized false fraction and an expected error rate, works through the adjustment algorithms, and separates dependence among tests from invalidity of the individual tests.

What you’ll learn

• Define the analysis target: distinguish feature contrasts, marginal and adjusted associations, observation-indexed summaries, and omnibus differences.
• Assess the evidence for the stated family: match uncertainty, dependence, resampling and multiplicity to the target, and limit the biological interpretation to what is identified.

Use the pages relevant to your research problem. Their checkpoints are completion targets for that topic; additional practice is optional. The group’s published manuscripts supply research case studies, while the self-contained numerical examples make particular arguments reproducible.