Start With the Biological Measurement Target
Choosing omics layers is not simply a matter of collecting the largest number of assay types. The useful question is: what biological quantities must be observed to address the research question? A layer is valuable when its assay measures a relevant quantity on suitable specimens, and when that measurement can be linked to the other selected layers for the subjects and sampling occasions of interest.
The introductory material on vertical integration distinguishes combining different molecular layers from combining cohorts that measure the same layer. Here the focus is narrower: deciding which layers belong in a proposed vertical integration. For example, a study of a signaling pathway might need measurements of RNA abundance, protein abundance, phosphorylation, and downstream metabolites. That list is a hypothesis about what must be measured, not a rule that every study should include all four.
An assay measures its specified molecular quantities; it does not, by itself, measure an entire biological process or establish a causal mechanism. A transcript measurement can inform a question about RNA abundance, but it is not a direct measurement of protein abundance. Likewise, phosphorylation measurements concern a protein's modification state, while an abundance assay may not resolve that state. Mapping the question to targets therefore prevents a common design error: treating convenient available data as if they directly answered the biological question.
A Practical Selection Framework
For each proposed layer, record the target, the specimen and sampling occasion required, and whether measurements can be linked to the same subjects as the other layers. This is a design audit, not a statistical test. It makes explicit which scientific requirements are met and where a proposed combination is infeasible.
Write down the molecular quantities needed to address the question. Separate directly measured quantities from quantities that would only be inferred indirectly.
For each assay, state what it measures, its specimen requirements, and whether it captures the relevant biological state. Do not count an indirect proxy as an equivalent direct measurement.
Ask whether each added layer contributes a distinct, question-relevant target. If two layers mainly repeat information, justify the added layer by a specific need, such as distinguishing abundance from modification state.
Verify that the selected measurements can be associated with the same subjects and compatible biological specimens and time points. Count unique subjects with all required layers, not the total number of assay records.
State which targets the selected combination covers, which it omits, and what conclusions the resulting measurements can and cannot support.
A target-by-layer table makes the mapping inspectable. “Direct” should mean that the measurement is of the target itself under the assay's stated measurement definition; it does not imply error-free measurement. “Partial” can indicate that an assay informs only one aspect of a broader target or requires additional assumptions to support an interpretation.
Worked Example: Matching Layers to a Signaling Question
Hypothetical study. A research group asks whether a treatment-associated signaling response is accompanied by changes in RNA, protein modification, and small-molecule abundance in tumor tissue. It is considering RNA sequencing, total proteomics, phosphoproteomics, and metabolomics.
| Candidate layer | Relevant target | Role for this question |
|---|---|---|
| RNA sequencing | Transcript abundance | Directly addresses the RNA component |
| Total proteomics | Protein abundance | Can add abundance information, but is not itself a phosphorylation measurement |
| Phosphoproteomics | Phosphorylation state of measured proteins | Directly addresses the modification component |
| Metabolomics | Small-molecule abundance | Directly addresses the metabolite component |
The question names three target classes: transcript abundance, phosphorylation state, and metabolite abundance. RNA sequencing, phosphoproteomics, and metabolomics therefore form a target-matched candidate combination. Total proteomics may be useful if the group also needs protein-abundance measurements, but the group should not include it on the claim that total protein abundance is interchangeable with phosphorylation state. The two targets are distinct.
This mapping does not show that changes in the three layers have been caused by the treatment or that one layer mediates changes in another. It specifies what the selected assays would measure. Before treating this as a feasible vertical design, the group must also establish that the chosen assays can be performed on compatible tumor specimens from the same subjects and the intended sampling occasions.
Shared Subjects and Specimen Compatibility
For a proposed combination of layers, joint analysis requires correctly linked measurements from the same observational units. Let \(S_k\) denote the set of unique subjects with usable measurements in layer \(k\), under the planned specimen and time-point definition. For a selected set of layers \(K\), the complete shared-subject set is
The number of shared subjects is \(n_{\mathrm{shared}}(K)=|S_{\mathrm{shared}}(K)|\). This count alone is not a measure of whether the study is adequate; it is a basic feasibility check. If adding a layer does not change the set of required subjects but some of its specimens are unavailable, the complete shared set can only stay the same size or become smaller. Formally, for layer sets \(K\subseteq L\), the intersection defining \(S_{\mathrm{shared}}(L)\) includes every set in the intersection for \(K\), and additional constraints, so
This is why an apparently attractive extra layer can reduce the number of subjects contributing to a complete joint analysis. Check the actual subject-to-assay map rather than assuming that cohort labels guarantee overlap. Also check specimen identity and biological timing: measurements from a subject's blood and tumor, or from baseline and follow-up, are not automatically measurements of the same biological state. They may still be useful, but their relationship must be part of the scientific design rather than silently treated as exact matching.
Worked Example: Calculating Overlap for Candidate Layers
Hypothetical data. A pilot inventory lists the following unique subjects with usable measurements at the intended tumor sampling occasion: RNA sequencing has subjects A, B, C, D, E, and F; phosphoproteomics has A, B, D, and G; metabolomics has B, C, D, and E.
For RNA sequencing and metabolomics, the shared set is \(S_{\mathrm{shared}}(\mathrm{RNA},\mathrm{metabolomics})=\{B,C,D,E\}\), so \(n_{\mathrm{shared}}=4\). For all three layers, intersect that set with the phosphoproteomics subjects:
Thus the three-layer combination has \(n_{\mathrm{shared}}=2\) subjects. The RNA-plus-metabolomics combination has four. If the research question requires phosphorylation measurements, two fully shared subjects in this pilot inventory reveal a joint-availability constraint that needs attention; counting all six RNA records or all four phosphoproteomics records as though they formed complete triples would be incorrect. The calculation says nothing by itself about whether two subjects are statistically sufficient, nor does it establish that the assays are biologically appropriate.
Assumptions, Diagnostics, and Failure Modes
The selection logic relies on several design assumptions. First, the research question has been translated into explicit targets. Second, each assay's measurement definition and specimen requirements are understood well enough to judge its relevance. Third, subject identifiers and specimen labels correctly link records to the intended biological units and occasions. Finally, the joint measurements refer to a biologically interpretable set of specimens; “same subject” alone does not guarantee “same tissue, state, or time.”
A practical audit should inspect the target mapping and the underlying availability table. Useful checks include whether subject identifiers are unique after resolving technical replicates, whether any layer is absent for a particular specimen type, whether sampling occasions differ systematically across layers, and whether assay failures leave only a narrow subset of subjects. Confirm that the overlap is counted at the planned unit of analysis. If multiple specimens or repeated occasions exist for one subject, they do not become additional independent subjects merely because they create additional rows.
Several warning signs indicate that the selection is going wrong:
- Layer names substitute for targets. Saying “we need RNA and protein” is incomplete until the study specifies which RNA and protein quantities matter.
- Proxy is mistaken for measurement. An association between two targets does not make a measurement of one a direct measurement of the other.
- Complementarity is assumed from technology. Two different assay platforms can still provide overlapping information, while two measurements in one broad assay family can address distinct targets.
- Availability is counted by records rather than subjects. Technical repeats or multiple specimens can inflate the apparent number of observations without increasing the number of uniquely shared subjects.
- Biological compatibility is overlooked. Measurements from different tissues or time points may answer different questions, even when they share a subject identifier.
If overlap is lower than expected, first determine whether it reflects a data-linkage error, an assay failure, a specimen constraint, or a genuine design choice. These have different remedies. A linkage error requires correction, not a biological interpretation. If the assays cannot use a common specimen, the study may need to narrow its layers, alter its specimen plan, or define a question that explicitly compares distinct specimens. Missingness patterns in omics data also require care: the subjects with complete measurements may not represent all enrolled subjects.
Comparing Design Alternatives
Selecting layers is a trade-off between relevance, feasibility, and interpretability. A narrow set of layers can support a more focused joint analysis and may preserve more complete shared subjects. A broader set can address more targets, but adds assay requirements and may shrink the complete intersection. The choice is not determined by a universal preference for more layers: it depends on which targets are essential to the stated question.
| Design choice | Potential advantage | Key limitation |
|---|---|---|
| Jointly measure a focused set of layers | Directly links selected targets across shared subjects | Omits targets not represented in the selected layers |
| Add further layers to the joint design | Can broaden the set of observed targets | Can increase specimen demands and reduce complete overlap |
| Measure one layer in a separate cohort | May make a difficult assay feasible without requiring every layer on each subject | Does not provide the same-subject measurements needed for direct subject-level cross-layer analysis |
| Use a narrower layer and follow up separately | Can prioritize a clearly defined target in the main study | Later measurements do not retroactively create joint observations in the original subjects |
A layer-selection decision should state both what the chosen combination enables and what it gives up. It cannot guarantee that the observed features will capture every aspect of a target, eliminate assay error, or establish causal direction. Nor does a high overlap count substitute for specifying the measurement target or confirming specimen compatibility. Those limitations should be described before interpreting cross-layer patterns.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A study asks whether a treatment changes transcript abundance and metabolite abundance in the same tissue. Which targets should be named before choosing assays, and what must be checked before joint analysis?
- Layer A has subjects \(\{1,2,3,4,5\}\), layer B has \(\{2,3,5,6\}\), and layer C has \(\{1,2,5,7\}\). Find the shared-subject set and its size for all three layers.
- Explain why total protein abundance is not automatically a substitute for phosphorylation-state measurement.
- A candidate fourth layer is scientifically relevant but is available for only some of the subjects with the first three layers. What is the design consequence to check, and does that fact alone prove the fourth layer should be excluded?
Solutions. (1) Name transcript abundance and metabolite abundance. Map each target to an appropriate assay, then verify that measurements can be linked for the same subjects, tissue, and intended sampling occasion. (2) The intersection is \(\{2,5\}\), so the shared-subject count is 2. (3) The targets differ: total abundance measures how much protein is observed, whereas phosphorylation measurement concerns a modification state. Protein abundance alone does not directly establish that state. (4) The complete shared set may shrink, so calculate the intersection with and without the fourth layer and assess the consequence for the scientific question. Relevance alone does not settle the decision; whether the lost overlap and added target are acceptable depends on the design aims and constraints.