Biology · Observation, Imaging & Experimental Methods

Model organisms, controlled perturbations and model limitations

An observed input and outcome vary together, but two candidate intermediate routes both reproduce the record. The correlation does not choose between them. A useful model or controlled perturbation helps frame an observation that could distinguish explanations, while keeping the question and biological context explicit.

Research Design, Data & Scientific Practice introduces questions, comparisons, uncertainty and independent units. Cells, Membranes & Organisation supplies the cellular boundaries and compartments behind localisation and counting. Gene Expression: RNA & Proteins explains why DNA and RNA can act as templates and why a measured reporter product is distinct from its input. Ordinary fractions, averages and the distance between planar points support the worked cases. Follow each supplied inventory and comparison first; the formulas summarise that reasoning rather than replacing the biological conditions.

Back to the Biology map

By the end, you can

Define why a biological model is useful, compare discriminating perturbation predictions with endpoint association, and retain the context and independence conditions of generalisation.

Choose a model for a question, not for universal resemblance

A model organism is a biological system used to investigate questions that may illuminate processes beyond that one system. Other models include selected cells, tissues or simplified molecular systems. Their value can come from accessible observation, defined backgrounds, established knowledge or a particular biological feature. No model reproduces every property of a different organism or environment.

A conserved component can support comparison without ensuring that all its partners, regulation or organismal consequences are conserved. Life stage, genetic background, sex, nutrition and environment can change an outcome even when a component has the same name. The relevant comparison is the mechanism and context needed for the question, rather than an unqualified ranking of which organism is most “like” another.

Background differences can matter within one species too. Genomic resources in the Hydra research system have been developed using different strains, a concrete reason to check whether records refer to comparable backgrounds. This is a methodological lesson about identity and context, not a claim that results from one Hydra strain never inform another. A useful model makes its transferable relationship and its unexamined differences clear.

A controlled change is interpreted through specificity and comparison

A perturbation changes a selected feature of a system to evaluate its response. A controlled comparison aims to distinguish that influence from other changes and observation effects. The biological or experimental unit, allocation of conditions, specificity, timing and measured outcome all help determine which claim the comparison can support. Merely observing a different endpoint after some change does not uniquely identify its cause.

The Boolean example supplies two hypotheses. In one, input A influences B and B influences outcome Y. In the other, A influences Z and Z influences Y. Both reproduce the observed endpoint relation: inactive A accompanies absent Y, while active A accompanies present Y. B and Z were not observed in those records, so matching A and Y does not establish their route.

The models make different predictions if B were selectively absent while A remained active. The B-route predicts absent Y; the Z-route predicts present Y. That is a discriminating logical comparison under the assigned equations. It does not demonstrate that a real biological change selectively affects B, that compensation is absent or that the chosen observation catches the relevant time. A response can constrain a candidate explanation without proving a unique direct pathway.

The same observed endpoints can fit different routes

A perturbation can test a discriminating prediction when its specificity, comparison and timing are justified. A response change alone need not establish a unique direct mechanism.

Read the diagram in words

Both hypotheses give Y absent when A is inactive and Y present when A is active. In the B-route model, selectively keeping B absent while A is active predicts Y absent. In the Z-route model it predicts Y present. Endpoint association alone does not identify which intermediate carries the influence.

Observed endpoint relation shared by both hypotheses
Observed input active	Observed outcome present
×	×
✓	✓

Hypothesis through B
Observed input active	Candidate intermediate B active	Candidate intermediate Z active	Observed outcome present
×	×	×	×
✓	✓	×	✓

Prediction with A active and B selectively absent
Observed input active	Candidate intermediate B active	Candidate intermediate Z active	Observed outcome present
✓	×	×	×

Hypothesised causal influence
Hypothesised causal influence	Hypothesised causal influence
Observed input A → Candidate intermediate B	Input influences B
Candidate intermediate B → Observed outcome Y	B influences the outcome

Hypothesis through Z
Observed input active	Candidate intermediate B active	Candidate intermediate Z active	Observed outcome present
×	×	×	×
✓	×	✓	✓

Prediction with A active and B selectively absent
Observed input active	Candidate intermediate B active	Candidate intermediate Z active	Observed outcome present
✓	×	✓	✓

Hypothesised causal influence
Hypothesised causal influence	Hypothesised causal influence
Observed input A → Candidate intermediate Z	Input influences Z
Candidate intermediate Z → Observed outcome Y	Z influences the outcome

Original Boolean mechanism hypotheses, not an experiment performed in cells or an operational intervention. A is the observed input and Y the observed outcome. B and Z are unobserved candidate intermediates. All functions and selective-B-absence assumptions are stipulated; no compensating route, off-target effect, timing, real gene or pathway identity supplied. Model predictions can distinguish these two candidates only if their assumptions apply.

A local result needs a separate transfer argument

A result belongs to the units and contexts observed, and generalisation asks how far it can inform other conditions. Independently assigned units within a model context can support a comparison within that context, while a much wider population or different species requires additional evidence. Hundreds of repeated observations of one background do not create hundreds of independently examined backgrounds.

The two-context record uses the same outcome scale and four assigned units per condition. In X, the reference mean is ten and the altered-condition mean six. In Y, both means are twenty. The descriptive contrasts differ. Their difference does not identify which contextual feature matters, and equality of the two recorded Y means does not prove a universally exact zero effect.

Complementary models can reveal which relationships persist and which depend on context. Disagreement can be informative rather than being discarded as a failed replication: it may expose a changed biological condition or measurement assumption. Broad explanation needs credible connections among specificity, mechanism, relevant independence and context. A theoretical candidate and an actual organism are therefore valuable forms of evidence with different limits.

A finding belongs to its biological context

Conserved components can make a model useful without making all biological contexts interchangeable. Generalisation needs evidence about the relevant mechanism and the differences that could change its outcome.

Read the diagram in words

X has means 10 and 6, contrast −4. Y has means 20 and 20, contrast 0. All four condition records contain four distinct assigned units. The difference between contrasts is −4. The equality of the two Y means in these supplied records does not prove an exact zero effect in every possible Y population.

Reference condition in X
Independent assigned unit identity	Assigned outcome units
Independent assigned unit 1 in this condition	9
Independent assigned unit 2 in this condition	10
Independent assigned unit 3 in this condition	10
Independent assigned unit 4 in this condition	11

Mean of four unit outcomes: 10

Altered condition in X
Independent assigned unit identity	Assigned outcome units
Independent assigned unit 1 in this condition	5
Independent assigned unit 2 in this condition	6
Independent assigned unit 3 in this condition	6
Independent assigned unit 4 in this condition	7

Mean of four unit outcomes: 6

Altered mean minus reference mean: -4

Reference condition in Y
Independent assigned unit identity	Assigned outcome units
Independent assigned unit 1 in this condition	19
Independent assigned unit 2 in this condition	20
Independent assigned unit 3 in this condition	20
Independent assigned unit 4 in this condition	21

Mean of four unit outcomes: 20

Altered condition in Y
Independent assigned unit identity	Assigned outcome units
Independent assigned unit 1 in this condition	19
Independent assigned unit 2 in this condition	20
Independent assigned unit 3 in this condition	20
Independent assigned unit 4 in this condition	21

Mean of four unit outcomes: 20

Altered mean minus reference mean: 0

Difference between the two descriptive contrasts: -4

Δ = ȳ₁ − ȳ₀

Original independently allocated unit records within each of two fictional biological contexts. All outcomes use the same assigned scale; reference and altered conditions are conceptual categories without a gene, organism, dose, operating procedure or actual experiment. Means are descriptive, with no fitted uncertainty, p-value, zero-population-effect proof or direct mechanism. Context includes background and environment; the supplied differences do not identify which contextual feature matters.

Worked example

Use predictions without inventing an experiment or a universal effect

In the first hypothesis, A active makes B active and then Y present. In the second it makes Z active and then Y present. With only A and Y observed, both hypotheses fit the two supplied endpoint states. No intermediate identity is selected by that fit.

The stipulated selective-B-absence state gives first-hypothesis outcome absent and second-hypothesis outcome present. Testing those candidates would require justified selectivity and observation conditions in the actual system. The truth table is a conceptual model prediction, rather than a performed intervention or a set of instructions for making one.

For X, the four reference values 9, 10, 10, 11 average to ten; altered values 5, 6, 6, 7 average to six. Its contrast is −4. Both Y conditions have 19, 20, 20, 21, mean twenty and contrast zero. The difference of descriptive contrasts is −4; no uncertainty estimate or proof of transfer to another context was supplied.

Check your understanding

Why do the shared A/Y records fail to establish whether B or Z carries the influence?
Show reasoning
Both candidate routes reproduce the same inactive/absent and active/present endpoints. The intermediates were not observed, and agreement at the endpoints leaves that part of the mechanism unresolved. A discriminating prediction needs a further observation whose interpretation is supported by the model’s conditions.
What limits the inference from the selective-B-absence truth table to an actual biological system?
Show reasoning
The example stipulates selective change, fixed equations and no compensation or observation delay. A real comparison would need evidence for those relevant conditions and appropriate independent units. Its outcome could reflect other affected features or timing, so the theoretical prediction alone does not establish a direct real pathway.
Do the two equal recorded Y means prove that the altered condition has no effect in every Y population?
Show reasoning
No. They are equal in this supplied finite record. The example provides no uncertainty estimate, complete population or evidence across all backgrounds and environments. The context comparison is descriptive and cannot justify an exact universal no-effect claim or identify the contextual mechanism.

What to carry forward

Use models and perturbations to compare explicit explanations, then justify specificity, independent units and context before generalising.

References

Resources supporting this lesson’s explanations and examples.