Data card — synthetic gene x disease

How to load

set.seed(123)
n <- 200
p <- 10
gene_names <- paste0("Gene", 1:p)
genes_data <- data.frame(matrix(rnorm(n * p), nrow = n))
colnames(genes_data) <- gene_names
disease_presence <- ifelse(
  0.5 * genes_data$Gene1 + 0.8 * genes_data$Gene3 - 0.3 * genes_data$Gene4 -
    genes_data$Gene1 * genes_data$Gene3 * 2 + rnorm(n, sd = 0.5) > 0,
  "Present", "Absent"
)
bio_data <- data.frame(genes_data, disease_presence = factor(disease_presence))

Where we use this

Caveats

  • Simulated — coefficients and interaction are for teaching, not biology.
  • Instructor ground truth: Gene1, Gene3, Gene4 (and Gene1 x Gene3 interaction) drive disease; other genes are noise.

Data card

1. Source

Simulated in R (set.seed(123)). Same story as logistic-regression-gene-disease.Rmd and the Day 1 data card example (archived).

2. Outcome

  • disease_presence: factor with levels Absent / Present (binary classification).

3. Predictors

Ten columns Gene1–Gene10, i.i.d. standard normal before the outcome is generated.

Gene Role
Gene1, Gene3, Gene4 Causal (linear terms + Gene1 x Gene3 interaction)
Gene2, Gene5–Gene10 Noise for classification

4. Sample size

n = 200. No missing values in the simulation.

5. Leakage risks

  • In-sample accuracy or AUC on the same rows used to tune or select features is optimistic.
  • Choosing hyperparameters (tree cp, lasso λ) on the full table before a holdout leaks information.
  • Day 2 fix: holdout test set; cross-validation inside training; microbiome labs add grouped resampling.

6. Subgroups

Not applicable in simulation.

Back to all datasets · Lab exercises