Biology · Biology in Health, Food & the Environment

From model systems to human evidence and translational limits

Two fictional groups have outcome frequencies of sixteen and thirty-four per hundred. Within each recorded baseline context, however, their outcome fractions are identical. A large-looking aggregate difference can arise from who is represented in the groups rather than from the factor named on their labels.

Research Design, Data & Scientific Practice introduces independent units, assignment, sampling and comparisons. Heredity, Variation & Evolution distinguishes inherited variants from traits and individual responses from population change. Energy, Enzymes & Metabolism introduces cellular reactions and material transformations. Their English chapters are available in this working manuscript; publication remains pending. The examples here use small graphs, fractions, sums and stated time windows, with each convention explained before use.

Back to the Biology map

By the end, you can

Distinguish mechanistic evidence, human outcome evidence and population representation, and recognise how context mixtures can create an aggregate difference.

A model answers a selected question

A research organism or model system is chosen to investigate a biological question. A cell system can make a molecular relationship easier to study; an intact organism can reveal interactions among tissues. These systems represent selected features. Their useful similarities to humans do not erase differences in physiology, development, context or the outcome measured.

The evidence bridge distinguishes three targets: a molecular activity in a model system, a response in a model organism and an outcome in people. Evidence at the first target can support a mechanism without establishing the third target. A change in a molecular readout is not numerically interchangeable with a change in a person's outcome.

Translation asks whether the relevant relationships carry across these settings. It includes whether the model represents the human condition and whether comparisons are reliable. Discordance can reflect bias or a mismatch between model and target. There is no single failure percentage that follows from calling something a model, and no automatic guarantee from a conserved mechanism.

Different evidence targets along a translational question

A useful mechanism can inform a human question while leaving the size and distribution of human outcomes unresolved.

Read the diagram in words

There are three separate targets: molecular activity in a model system, a response in a model organism, and a defined outcome in people. One link asks about operation in the organismal context; the other asks about operation and outcome in people. Neither link guarantees agreement.

Scope question
Evidence target (1)	Evidence target (2)	Scope question
Model-system molecular activity	Model-organism response	Does the relationship operate in this organismal context?
Model-organism response	Defined human outcome	Does it operate and affect this human outcome?

Correspondence to examine

Original conceptual correspondence model; no empirical treatment, effect size, success rate or sequential requirement is supplied.

The composition of a group can change its average

In the fictional comparison, a lower-baseline context has one recorded outcome per ten people and a higher-baseline context has four per ten. Both groups have these same fractions within each context. Group Mixed One contains eighty people in the lower context and twenty in the higher; Mixed Two contains twenty and eighty.

Mixed One therefore has eight plus eight outcomes, or sixteen among one hundred people. Mixed Two has two plus thirty-two, or thirty-four among one hundred. The contexts are supplied as complete records, not inferred from the totals. The eighteen-per-hundred difference arises despite identical within-context fractions.

An aggregate association may therefore be entangled with differences in baseline composition. Context-adjusted comparisons address a different question from an unadjusted total. They also rely on knowing the relevant contexts and their relation to the outcome. Adjustment for one recorded factor does not prove that all other sources of bias have been removed.

Different group weights, identical context fractions

The eighteen-percentage-point aggregate contrast is reproduced entirely by the supplied context weights. It does not identify a causal group-label effect.

Read the diagram in words

Mixed One has eighty lower-context people with eight outcomes and twenty higher-context people with eight: sixteen of one hundred. Mixed Two has twenty with two and eighty with thirty-two: thirty-four of one hundred. Each within-context fraction matches. The separate equal-weight accounts each have fifty with five and fifty with twenty: twenty-five of one hundred.

Mixed One
Baseline context	People	Recorded outcomes	Outcomes within context
Lower-baseline context	80	8	0.1
Higher-baseline context	20	8	0.4

Outcomes among all group members
Comparison group	People	Recorded outcomes	Outcomes among all group members
Mixed One	100	16	0.16

Mixed Two
Baseline context	People	Recorded outcomes	Outcomes within context
Lower-baseline context	20	2	0.1
Higher-baseline context	80	32	0.4

Outcomes among all group members
Comparison group	People	Recorded outcomes	Outcomes among all group members
Mixed Two	100	34	0.34

Equal-weight One
Baseline context	People	Recorded outcomes	Outcomes within context
Lower-baseline context	50	5	0.1
Higher-baseline context	50	20	0.4

Outcomes among all group members
Comparison group	People	Recorded outcomes	Outcomes among all group members
Equal-weight One	100	25	0.25

Equal-weight Two
Baseline context	People	Recorded outcomes	Outcomes within context
Lower-baseline context	50	5	0.1
Higher-baseline context	50	20	0.4

Outcomes among all group members
Comparison group	People	Recorded outcomes	Outcomes among all group members
Equal-weight Two	100	25	0.25

Outcomes among all group members: 0.34 − 0.16 = 0.18

Original complete fictional records and separate equal-weight accounts under the same stipulated fractions. No actual random assignment, intervention, sampling uncertainty or population transportability is established.

Internal comparison and wider relevance are separate

Random assignment allocates study participants or units to comparison conditions by chance. It can make a causal comparison more credible by balancing background factors in expectation, although any finite assignment can remain unbalanced. Representative sampling instead concerns how the study units relate to the population of interest. One operation does not automatically accomplish the other.

In a purely arithmetic comparison with fifty people from each context, the stipulated fractions give five plus twenty outcomes, or twenty-five per hundred, in each group. These are supplied accounts under an unchanged rule. They illustrate equal weighting; they are not results from a randomized human study and do not establish an intervention's effect.

Human evidence must also specify the outcome, follow-up, missing information and people to whom a conclusion applies. A reliable comparison in one set of participants may leave uncertainty about another context. Agreement among mechanistic and human findings can strengthen an explanation, while disagreement identifies assumptions to examine rather than a reason to ignore either type of evidence.

Worked example

Rebuild the aggregate difference from its parts

The lower-context fractions are eight divided by eighty and two divided by twenty. Both equal one tenth. The higher-context fractions are eight divided by twenty and thirty-two divided by eighty. Both equal four tenths. There is no within-context difference in this stipulated record.

Summing gives sixteen divided by one hundred for Mixed One and thirty-four divided by one hundred for Mixed Two. Subtracting gives eighteen percentage points. This is an aggregate descriptive contrast between differently composed groups; it does not identify the effect of their group labels.

Giving both contexts weight one half produces (one tenth plus four tenths) divided by two, or one quarter. The two groups then share the same weighted value. That calculation answers what the fractions would average under equal context weights, not whether a real application is beneficial or whether these two contexts capture every relevant difference.

Check your understanding

Mixed Two has more outcomes overall. Must its group label cause an increase in the outcome?
Show reasoning
No. The displayed record has equal within-context fractions and different context weights. Those weights account for the aggregate contrast. A group-label effect cannot be identified from that contrast alone.
Does random assignment make a study representative of every population?
Show reasoning
No. Assignment concerns comparison conditions among the included units; sampling and participation concern which units are included. Credible internal comparisons can still have limited relevance to people or settings that were poorly represented.
A model-system molecular response matches a proposed mechanism. What remains open about a human application?
Show reasoning
Whether that relationship operates in the relevant human context, whether it changes the defined human outcome, and how benefits or burdens are distributed. The molecular response supplies mechanistic evidence at its own scale, not a complete human outcome estimate.

What to carry forward

A model, a causal comparison and a population conclusion have different evidence requirements. Preserve the target outcome and the composition of the groups.

References

Resources supporting this lesson’s explanations and examples.