Safety, privacy and evidence boundary
Use fictional organisms and de-identified/public datasets. Do not infer intelligence, personality, worth, disease, ancestry, race, or family traits from genetics. Complex-trait models are population-level scientific tools, not labels for individuals.
Explicit science instruction
Complex-trait studies often report correlations among genotype scores, environments, and phenotypes. Correlation measures association; it does not by itself establish direction, mechanism, or absence of confounding.
Prediction can be useful even when causal understanding is incomplete, but predictive accuracy must be evaluated in the population and conditions where the model will be used.
Worked example with reasoning
A fictional polygenic score correlates r = 0.45 with plant height in one greenhouse. That indicates moderate association in that sample. It does not mean genes explain 45% of an individual's height, nor does it prove the same prediction in another environment.
Guided question
Why can a model predict reasonably well while still being a poor causal explanation?
Independent work
- Interpret three fictional scatterplots.
- Separate association, prediction, and causation statements.
- Name two confounders.
- Explain why validation in a new sample matters.
Investigation / Data Lab
Reaction-norm dataset investigation
Question: Does environmental condition change genotype effects in the supplied fictional organism dataset?
- Graph phenotype for at least two genotypes across three environmental conditions.
- Calculate the phenotype difference between genotypes within each environment.
- Describe whether the response lines are roughly parallel, diverging, converging, or crossing.
- State what pattern is consistent with gene–environment interaction and what the graph cannot prove about mechanism.
- Propose one controlled follow-up measurement and one confounder to avoid.
Evidence standard: keep the raw table, graph, difference calculations, and a bounded causal claim.
Misconception check
A correlation coefficient is not a percentage of causation and should not be converted into personal destiny.
Formative understanding check
What does r = 0 mean?
Worked response
No linear association in the sample; nonlinear relationships may still exist.
Open-book formative practice only; not an official mastery attempt or grade.
Mentor evidence conference
Ask Archaemenes to challenge the model, quantitative reasoning, and limits without inferring personal genetic or medical information.
Open ArchaemenesTeacher guidance
Foundation: use sentence frames for association only. Core: evaluate predictive claims and limits. Extended: compare training and validation samples and discuss transportability.
Preserve learning evidence
Keep the model, data interpretation, correction note, and limitation in the matching Week 22 evidence field.