What Does One Feature Represent?
In multi-omics data, a feature is not just a name attached to a numerical value. It refers to a particular biological or assay-defined unit: for example, a genomic variant locus, an RNA transcript, a methylation site, or a peptide. Two features can be related to the same broader biological entity without representing the same unit or measurement. Keeping these distinctions clear is essential when describing results or deciding whether features can sensibly be compared or combined.
Here, feature granularity means the level of biological or assay-defined detail represented by one feature. It is distinct from the biological sample or subject from which an observation is obtained. A subject may contribute measurements at many variants, transcripts, methylation sites, peptides, or peaks; those are different feature units measured on that subject. The measured values at those features also need not have the same meaning or scale.
The units in the layers considered here differ in important ways. A variant feature refers to a genomic locus at which alleles differ among individuals. A gene feature refers to a gene-level unit, while a transcript feature refers to a particular RNA transcript produced from a gene. A methylation-site feature refers to a measured site, commonly a particular cytosine in a genomic context. A peptide feature refers to a peptide sequence observed as protein evidence; a protein feature refers to a protein-level target or estimate. A metabolite-peak feature refers to a signal detected by an analytical procedure, not necessarily to a uniquely identified compound.
These distinctions build on the earlier material on Measurement Targets Across Omics Assays and the assay-specific material on genomic variants, RNA sequencing, epigenomic tables, proteomic tables, and metabolomic tables. The aim here is to compare the units across layers, not to treat all measurements as if they were the same kind of quantity.
Related Units, Different Information
| Feature level | What one feature represents | What it should not be mistaken for |
|---|---|---|
| Variant | A genomic locus at which a genotype or allele dosage is recorded | A gene’s expression or a summary of all variation in that gene |
| Gene | A gene-level unit or summary | A particular transcript or a single genomic variant |
| Transcript | A particular RNA transcript | The whole gene’s activity in every transcript-related sense |
| Methylation site | A particular assayed site, often a CpG site | A gene-wide methylation state |
| Peptide | A peptide sequence observed as protein evidence | A complete protein molecule or an unambiguous protein amount |
| Protein | A protein-level target or estimate | A single peptide observation |
| Metabolite peak | A detected signal at a measured peak | A confirmed, uniquely quantified compound |
The table is a guide to the unit, not a guarantee that every assay produces every type of feature. Within each layer, actual measurement and summarization procedures matter. For example, a gene-level RNA value may be computed from transcript-level information, but its interpretation depends on the procedure and question. A peptide can provide evidence for a protein, but a peptide observed in more than one protein context may not identify which protein generated the signal. A peak can be associated with a candidate compound without being uniquely attributable to it.
Granularity matters because finer features can preserve distinctions that a broader summary removes. Conversely, a broader summary may be more directly suited to a question that concerns the broader unit. Neither level is automatically superior. The correct choice depends on the biological question and what the assay supports.
Aggregation Changes the Target
Suppose a gene has several transcript features, with a value for each transcript in a sample. A gene-level summary may add transcript values, average them, or use another defined procedure. Even when the chosen summary is a sum, the total does not retain the transcript composition. This is a general consequence of many-to-one aggregation, not a special property of one assay.
The sum is meaningful as a gene-level summary only under conditions that make its component values commensurate and the aggregation appropriate. For example, the transcript values must refer to the same sample, and the analysis must have a defensible reason to combine those transcript units. Even then, equal totals can conceal different transcript profiles.
Worked Example: A Gene Total Hides Transcript Composition
Consider a hypothetical gene with two transcript features, T1 and T2. For sample A, suppose the measured transcript values are 70 and 30 units. For sample B, suppose they are 40 and 60 units. Assume the values are on a common scale, refer to the same gene and sample within each profile, and are being summed solely to illustrate a gene-level total.
For sample A, the total is \(70 + 30 = 100\) units. For sample B, it is \(40 + 60 = 100\) units. Thus, both samples have the same total under this summary, although the transcript profiles differ: T1 is higher in sample A by \(70 - 40 = 30\) units, while T2 is higher in sample B by \(60 - 30 = 30\) units.
The gene total answers a different question from the two transcript values. It summarizes the combined value under the stated rule, but cannot reveal which transcript contributed more. It would be incorrect to conclude from equal totals that the transcript-level profiles are equal.
This example also shows why a transcript should not simply be treated as a gene. A gene can have multiple transcript features, and a gene-level summary can erase variation among them. In the other direction, a transcript-level observation is not automatically a complete description of all gene-level activity.
Examples Across Molecular Layers
Worked Example: Variant Loci Are Not Gene-Level Expression Features
Suppose a hypothetical study records allele dosage at two variant loci, L1 and L2, in the same gene for two subjects. The dosage values are 0 and 2 for subject A, and 1 and 0 for subject B. The values describe the recorded dosages at the respective loci; they are not expression measurements.
The sum of the two recorded dosages is \(0 + 2 = 2\) for subject A and \(1 + 0 = 1\) for subject B. That arithmetic gives a sum of the two locus-specific dosage values, but it does not turn the result into a gene-expression measure or a general measure of the gene’s state. It also conceals which locus contributed the dosage: subject A’s value comes entirely from L2, whereas subject B’s comes from L1.
The assumptions for this calculation are only that the two dosage values are recorded for the same subject and that their numerical sum is being used as an arithmetic illustration. A scientific interpretation of that sum would require a separate justification. The proper feature-level descriptions remain “dosage at L1” and “dosage at L2.”
Worked Example: Peptide Evidence Does Not Automatically Equal Protein Abundance
Imagine a hypothetical proteomic run in which peptide P1, observed as evidence for protein A, has an intensity of 120 units. Peptide P2 has an intensity of 80 units and is compatible with protein A and protein B. Suppose, only for illustration, that someone adds both values and reports \(120 + 80 = 200\) units as the abundance of protein A.
The arithmetic is correct, but the protein-level interpretation is not established by it. P2 is shared evidence, so its 80 units cannot be assigned uniquely to protein A from the information given. Assigning all of it to A assumes the signal came from A; assigning none assumes it came from B. Neither allocation follows from these two observations alone.
The values are hypothetical peptide signals from the same run, but that alone does not justify summing them as a protein abundance estimate. A protein-level estimate requires an explicit procedure that handles the relationship between peptide evidence and candidate proteins. The feature units are peptide P1, peptide P2, protein A, and protein B; they are related, but they are not the same.
Worked Example: Multiple Peaks Do Not Necessarily Mean Multiple Compounds
Suppose a hypothetical sample has two detected peaks, with signal values of 500 and 300 instrument units. Both are candidate signals for the same metabolite, perhaps because the analytical method can produce more than one signal associated with a compound. Their arithmetic sum is \(500 + 300 = 800\) instrument units.
That sum is a sum of two peak signals, not automatically a measured concentration of the metabolite. To interpret 800 as a compound-level quantity would require evidence that both peaks represent that compound and a justified rule for combining their signals. Without that support, the features remain two detected peaks. Even if both are associated with one compound, that association does not establish that their values can be added as a compound abundance.
The assumption needed for the arithmetic is merely that both recorded signal values are available in the same hypothetical sample. No assumption about compound identity or the meaning of the combined value has been supplied, so the calculation cannot support a compound-level conclusion.
Choosing and Checking a Feature Level
A useful check is to state what one feature means before describing a pattern or comparing values. For a variant, name the locus; for an RNA measurement, distinguish a gene-level unit from a transcript; for methylation, specify the site rather than implying a gene-wide state; for proteomics, distinguish peptide evidence from a protein-level estimate; and for metabolomics, distinguish a detected peak from a compound. These descriptions prevent an interpretation from silently shifting to a broader or narrower unit.
When a summary is used, make the change of unit explicit. For example, “the transcript values were summed to form a gene-level summary” is more informative than calling the sum “expression” without specifying its construction. When a result is reported at the original feature level, do not describe it as a broader biological conclusion unless the relationship is supported.
State whether the observation refers to a locus, gene, transcript, methylation site, peptide, protein, or detected peak.
Describe what the assay’s value represents at that unit, without replacing it with a related but different biological quantity.
Identify the component features, the aggregation rule, and whether that rule supports the intended broader interpretation.
Phrase the result at the feature level actually measured or summarized; do not silently generalize it to another level.
Diagnostics for a granularity problem are often conceptual rather than numerical. Ask whether the feature definition and conclusion name the same unit; whether one feature could correspond to multiple broader entities or vice versa; and whether a summary has discarded distinctions relevant to the question. If an interpretation requires assuming that a shared peptide belongs to only one protein, or that a peak uniquely represents a compound, check whether that assumption is supported rather than implicit.
A second diagnostic is to compare what would be retained under plausible choices of granularity. Transcript-level analysis can preserve transcript differences that a gene total hides. Site-level methylation measurements can differ within a gene, so one site does not automatically stand for the whole gene. A peptide-level signal may be informative without being an unambiguous protein estimate. A peak can be useful as a measured feature even when compound identity is uncertain. In each case, changing the unit can change the question answered.
Common Mistakes and Limitations
- Wrong: “These two features are related to the same gene, so they are interchangeable.” Correction: A shared broader context does not make a locus, transcript, methylation site, or peptide the same measurement unit.
- Wrong: “The sum of finer-scale values is automatically the broader-scale value.” Correction: The arithmetic may be valid while the biological interpretation is not. State and justify the aggregation rule.
- Wrong: “A peak is a compound,” or “a peptide is a protein.” Correction: A peak is a detected signal and a peptide is protein evidence; the relationship to a compound or protein requires appropriate support.
- Wrong: “A gene-level result tells us which transcript changed.” Correction: A gene-level summary can conceal transcript composition. Examine the transcript-level values if that distinction matters.
There is no universal conversion that makes measurements at these different levels equivalent. Assay design, feature definitions, and the biological question constrain the available interpretation. In particular, a mathematical transformation cannot by itself establish biological identity or justify a cross-layer comparison. The practical goal is not to force all layers into one granularity, but to preserve the unit being measured and make any change of unit explicit.
Check Your Understanding
Answer each question by identifying the feature unit and stating what conclusion the data do, or do not, support.
- A gene has three transcript features. What information can a gene-level sum lose?
- A measured peptide is compatible with two proteins. Can its intensity alone be assigned uniquely to either protein? Why or why not?
- Two peaks have signal values of 40 and 60 instrument units and are both candidate signals for one compound. What is their arithmetic sum, and what does that sum establish?
- A study records dosage at two variant loci in one gene. Does the sum of those dosages measure gene expression? Explain.
Solutions. 1. The sum retains the total under the chosen rule but loses how that total is distributed among the three transcripts; profiles with different transcript values can have the same sum. 2. No. Compatibility with two proteins means the observation alone does not establish which protein generated the signal, so unique assignment needs additional evidence or an explicit estimation procedure. 3. The arithmetic sum is \(40 + 60 = 100\) instrument units. It establishes only the sum of the two peak signals; it does not establish a compound-level quantity without supported identity and a justified combination rule. 4. No. The sum is an arithmetic summary of dosage values at the two loci. It is not a gene-expression measurement because the feature units and measurement targets differ.