Skip to content

Module 3: Diagnosing design problems in omics

Learning objectives

  • Apply diagnostic thinking across different data types and experimental design elements
  • Assess whether experimental limitations can be addressed through analysis or require redesign

In Module 1 and Module 2, we looked at the omics layers, the key design considerations, and the trade-offs between accuracy, interpretability, power, cost, and generalisability. Module 3 brings those ideas together by asking how to diagnose whether a study is asking the right question, measuring the right thing, and designing around the right constraints.

To support tying together some of the earlier cocepts we have made a mental model checklist

Criterion Question Considerations
Molecular model fit Is the chosen molecular layer appropriate for the biological question? Which molecular layer best matches the biology being studied; what each omics layer captures and what it cannot tell you; whether the measurement is close to the biology of interest rather than a downstream or indirect readout. See 2.1.1 Choosing a measurement platform.
Interpretability Will the result mean what we think it means? Confounding and batch effects; blocking and randomisation; normalisation appropriate to the data; tissue or cell composition; positive and negative controls; statistical versus biological significance. See 2.1.2 Confounding: when a variable travels with your groups and 2.1.3 Measurement reliability and metadata.
Metadata completeness Is the study documented well enough for others to interpret and reproduce it? Sample metadata such as age, sex, tissue, disease status, collection site, timing, and processing conditions; batch identifiers; storage, handling, and instrument metadata; knowing which variables are missing and how those gaps affect interpretation. See 2.1.3 Measurement reliability and metadata.
Signal detection Can the design detect the effect we care about? Statistical power given the correct error model; multiple-testing burden; the unit of replication; missingness or dropout reducing effective sample size; whether the study was designed around the size of effect it intends to detect. See 2.2.1 Sample size and statistical power.
Resource use Is the budget being used effectively? Sequencing depth versus breadth; technical versus biological replication; budgeting for QC attrition; platform cost versus the needs of the research question. See 2.2.2 Design decision costs.
Generalisability Will the results hold outside our dataset? Cohort representativeness versus the target population; whether results hold across different omics data types; reference and annotation bias; whether independent validation is possible. See 2.3.1 Generalisability: who and what the findings apply to.
Design coherence Are the scientific question, cohort, variables, and platform aligned before data are generated? Clear biological question; explicit hypothesis; well-defined comparison groups; relevant covariates and confounders; a platform that can answer the question within the study's cost and time constraints; a realistic statement of what the design can and cannot support. See 2.4.1 Design decisions in practice.