Experimental design, metadata and reproducibility
A treatment is an independent variable; an outcome is a dependent variable.
Loading video…
A treatment is an independent variable; an outcome is a dependent variable. Controls, biological replicates and collection metadata help distinguish a treatment effect from confounding. A technical repeat of one sample is not a new biological replicate. Track provenance and missingness before analysis; later, keep linked samples together across train and test splits.
Worked example
If all control samples were collected Monday and all treated samples Friday, batch and treatment are confounded. Rebalance collection or state that the effect cannot be separated.