Fig. 1

Conceptual schema of eNetXplorer. (a) One or more datasets can be aggregated into an N×p input matrix. Depending on the GLM of interest, the response is a numeric vector (linear), 2-class factor (binomial) or multi-class factor (multinomial). (b) Ridge and lasso implement different regularization penalty terms, which are tuned by the regularization parameter λ. (c) The elastic net introduces the mixing parameter α as a continuous tuner from ridge to lasso. (d) eNetXplorer generates a null-model ensemble via random permutations of the response (d₁), which allows the quantitative exploration of elastic net families by assessing the statistical significance of each model (d₂), the feature-level significance within each model (d₃) and across the entire elastic net family (d₄)