An RNA spike-in provides an external reference, but the unit attached to that reference decides what the comparison means. The same added amount per cell, per tube or per microgram of extracted RNA can support different conclusions. A table labelled “spike-in normalised” is incomplete until that experimental denominator is known.
This guide follows one question: did the amount of a particular RNA per cell change? We build a small measurement model, work a cell-count example and then change how the controls are added. The calculations are original idealised examples. They explain the logic of an experiment; they do not establish that any specific control mixture perfectly reproduces endogenous RNA behaviour.
eduKateSG · The Gold Standard of Bioinformatics
Choose your reading route
Define the reference and biological unit · Chapters 1–2. Build the quantities before using a formula.
Follow cell counts and control addition · Chapters 3–5. Follow original examples, assumptions and changed conditions.
Test the denominator and continue · Chapters 6–8. Test transfer, answer common questions and follow connected reading.
Full chapter index · Practice and evidence checks · Return to Bioinformatics
Chapter index
Define the reference and biological unit · Chapters 1–2
Follow cell counts and control addition · Chapters 3–5
Test the denominator and continue · Chapters 6–8
CHAPTER 1 OF 8 · Define the reference and biological unit
1. Choose the biological unit before choosing a normalisation
Suppose two cultures contain the same proportions of transcripts, but one has twice as much measurable RNA per cell. Sequencing a fixed number of fragments from each library can produce very similar proportions. A composition can remain unchanged while the amount behind it grows.
That observation is not a failure of relative abundance. Relative and absolute-scale questions are different. “What share of the measured transcript pool belongs to gene A?” asks about composition. “How much gene A RNA is present per cell?” asks for a biological scale outside that composition.
Start the experimental description with the intended unit: per cell, per nucleus, per tissue mass, per culture volume or per mass of extracted RNA. These are not interchangeable. A tissue sample may change cell composition; cells may differ in size, viability or RNA content; a nuclear assay measures a different material from a whole-cell assay.
The Bioinformatics root connects biological units with metadata and computation. The narrower task here is to preserve a reference through sample preparation so a chosen unit remains interpretable after sequencing.
The NIST External RNA Controls Consortium overview describes external controls as tools for evaluating expression measurements. Their presence does not, by itself, determine the biological unit of every result.
CHAPTER 2 OF 8 · Define the reference and biological unit
2. Build the ratio and expose its assumptions
Let n be the number of biological cells contributing to a sample, r the amount of target RNA per cell and S the amount of external control added to the processed material. Let cᵣ and cₛ be the corresponding observed counts or count summaries.
In an ideal linear model, cᵣ = a × eᵣ × n × r and cₛ = a × eₛ × S. Here a represents a shared sampling scale and eᵣ/eₛ represents the relative response of the target and control through the stages they share.
Their ratio gives cᵣ/cₛ = (eᵣ/eₛ) × n × r/S. Rearranging yields r = (cᵣ/cₛ) × (S/n) × (eₛ/eᵣ).
For comparing two samples, a stable relative response can allow the response ratio to cancel. Without suitable calibration, however, multiplying by the known control amount does not automatically produce an accurate absolute molecule count for an unrelated endogenous transcript.
The equation reveals three distinct needs: the observed ratio, the control amount relative to the biological unit and evidence that the relative response is sufficiently stable. A spreadsheet can calculate the first. It cannot reconstruct missing cell counts or make a late-added control experience an earlier extraction loss.
Bioconductor’s normalisation chapter, section 2.5 states the key assumptions for per-cell spike-in scaling: a controlled addition and sufficiently comparable responses to technical biases. Those assumptions should be checked against the actual protocol.
CHAPTER 3 OF 8 · Follow cell counts and control addition
3. Work an unequal-cell-count example
Consider two hypothetical bulk samples processed with the same amount of external reference. Use 10,000 control units per tube. Sample A contains 100 cells; sample B contains 200. The observed target-to-control count ratio is 1.0 in both.
| Sample | Cells n | Added control S | Target/control count ratio | Ratio × S/n |
|---|---|---|---|---|
| A | 100 | 10,000 units | 1.0 | 100 reference-scaled units per cell |
| B | 200 | 10,000 units | 1.0 | 50 reference-scaled units per cell |
The inferred per-cell fold change B/A is (1.0/1.0) × (10,000/10,000) × (100/200) = 0.5, assuming stable relative response and representative sampling.
The count ratio indicates the same target amount relative to the fixed control in each tube. In this model, twice as many cells share that amount in B. Dividing by the biological unit changes the interpretation.
Now change the design. Add 100 control units per cell, giving 10,000 units to A and 20,000 to B. If the target amount per cell is unchanged, both target/control ratios should again be equal. This time equal ratios do imply equal per-cell amounts under the model because S/n is deliberately constant.
The same-looking count result can therefore support different conclusions under different addition rules. Preserve the rule in the sample metadata, not only in a notebook sentence that disappears before analysis.
For exact molecule claims, additional calibration is needed. The phrase “reference-scaled units” keeps the ideal calculation separate from an assertion that every endogenous molecule has been counted without bias.
CHAPTER 4 OF 8 · Follow cell counts and control addition
4. Change the addition rule from cells to RNA mass
Suppose condition A contains 10 arbitrary RNA-mass units per cell and B contains 20. The fraction contributed by the target is unchanged. The target amount per cell has doubled, even though its share of the RNA pool is constant.
Take an equal extracted RNA mass from each condition and add the same control amount to those equal-mass aliquots. In the ideal model, the target/control ratios will be equal. Each tube contains the same target amount per unit of extracted RNA; B’s aliquot represents fewer original cells.
| Design choice | Reference held constant | Comparison directly supported under the model |
|---|---|---|
| Equal control per counted cell | Control amount/cell | Target amount per cell |
| Equal control per extracted RNA mass | Control amount/RNA mass | Target amount per RNA mass |
| Equal control per tube | Control amount/tube | Target amount per processed tube; cell counts are needed for a per-cell result |
If cell counts, extraction yields and aliquot fractions were recorded, some designs permit a justified conversion between units. If they were not, the per-cell scale may be unidentified. A normalisation method cannot infer it uniquely from proportions alone.
This is why a global shift can disappear after equal-RNA-mass sampling and relative scaling. No arithmetic error is required. The experiment has defined a unit that answers a different question.
That design may be appropriate when the scientific goal is composition per unit RNA. The correction is to align the conclusion with the design, rather than declaring one normalisation universally best.
CHAPTER 5 OF 8 · Follow cell counts and control addition
5. Follow what the control actually experienced
A synthetic RNA added after extraction can monitor aspects of library preparation and sequencing, but it did not experience cell disruption or RNA loss during extraction. A control added before those stages may cover more of the workflow, yet free synthetic RNA can still behave differently from RNA enclosed in a cell or associated with cellular structures.
Draw the chain explicitly: biological material → lysis → extraction → aliquoting → control addition → library preparation → sequencing → quantification. Move the control-addition point to its real position. Any stage upstream requires separate evidence or an explicit limitation.
Also inspect dose accuracy, mixing, degradation, sequence characteristics, concentration range and whether the control occupies enough counts for a stable estimate. A tiny control count has substantial sampling variability; an excessively large addition can consume sequencing capacity.
Use multiple suitable control transcripts to examine consistency. If all behave similarly, a shared scale explanation is more plausible. If responses vary strongly by concentration or sequence, a single scaling factor may conceal a problem. Agreement among controls still does not prove they track every endogenous transcript.
An unchanged housekeeping gene is an alternative anchor only if its stability is justified for the conditions. Selecting it because it gives the expected answer would make the argument circular.
The practical result is a bounded claim: a reference can help preserve a scale through the stages it represents, subject to tested response assumptions. It does not remove biological variability, sample-composition differences or the need for independent replicates.
CHAPTER 6 OF 8 · Test the denominator and continue
6. Test the reference in changed conditions
Case A: a different spike dose
A has 100 cells, 10,000 control units and a target/control ratio of 2. B has 100 cells, 20,000 control units and a ratio of 1. Did the target halve?
Under the ideal model, both estimates are 200 reference-scaled units per cell: 2 × 10,000/100 and 1 × 20,000/100. The count ratio halved because the external denominator doubled. Ignoring the dose record would reverse the interpretation.
Case B: the same per-cell amount, a weaker response
Both samples contain equal cell numbers and equal control doses. The target/control ratio in B is 0.8 of A. Suppose independent validation shows that the target-to-control response ratio in B is also 0.8 of A. The model then permits unchanged target amount per cell. A changed response and a changed biological amount can produce the same observed ratio.
This is not a correction to apply whenever a result looks inconvenient. The response difference needs evidence independent of the desired conclusion.
Case C: a missing cell count
The ratios and control doses are known, but B’s cell count is missing. You can compare target amount per processed tube under the assumptions. You cannot uniquely calculate the per-cell fold change. Report the supported unit and identify the missing denominator.
An analysis-ready record should include sample identity, biological unit, cell-count method where relevant, viability criteria, extracted amount, aliquot fraction, control identity and lot, control dose, addition stage, batch, response checks and independent replicate identity.
Keep original counts alongside derived values. A count-based statistical method requires an appropriate count input and its documented normalisation strategy; do not substitute a ratio table into it merely because both tables contain numbers.
CHAPTER 7 OF 8 · Test the denominator and continue
7. Questions about external references
Does a spike-in make RNA-seq absolute?
It can support calibrated or reference-scaled measurements when the design and response model justify them. A fixed addition alone does not establish unbiased absolute copies for every endogenous transcript.
Can equal control coverage be a useful target?
Yes, when the protocol and question justify scaling by the controls. Equalising coverage means removing variation relative to that external reference. It does not imply that total endogenous RNA should be equal.
Are per-cell and per-microgram results contradictory?
They may both be correct. If total RNA per cell changes, the denominators differ. State the unit before comparing the conclusions.
Can a control added after extraction measure extraction efficiency?
It cannot directly measure losses that occurred before it entered. An earlier control also needs evidence that its behaviour represents the biological material through extraction.
Do spike-ins replace replicates?
No. They provide measurement references. Independent biological replicates are needed to estimate biological variability and support generalisation beyond the particular samples.
CHAPTER 8 OF 8 · Test the denominator and continue
8. Connect reference design with the wider analysis
Understand the expression workflow
- The Gold Standard of Bioinformatics connects units, reference records, replication and reproducibility.
- RNA Sequencing, Count Normalisation and Differential Expression develops count models and relative scaling. Use this article when the missing question is the experimental unit behind an external reference.
- Independent Filtering and Why an Adjusted P-Value Can Be Missing explains a later stage of inference, after the input and design have been defined.
Check the reference assumptions
- Bioconductor: Normalisation, section 2.5, sets out the spike-in scaling assumptions.
- NIST: External RNA Controls Consortium explains the measurement-assurance purpose of external controls.
- Lun and colleagues: Assessing the reliability of spike-in normalisation investigates the approach and its assumptions.
The transferable habit is to write the denominator beside the result and follow it back to the experimental action that created it. Continue through Learning That Transfers into Life and How X Works to compare other systems in which the reference determines the claim.