Biomarker Discovery¶

GrAndMA includes two modules for identifying and validating biomarkers from omics data.

Discomarker¶

Discovers candidate biomarkers using regularised regression with SHAP-based feature importance and optional consensus clustering.

A preprocessed dataset
A sample metadata column to use as the outcome variable (binary class label or continuous value)

Loads the preprocessed feature matrix and merges the selected outcome column from sample metadata
Fits an elastic net model with cross-validation to select features predictive of the outcome
Computes SHAP values to rank features by their contribution to the model
Optionally runs consensus clustering across a range of k to identify sample subtypes

Parameter	Description
Outcome column	Metadata column to predict (binary or continuous)
Group column	Optional stratification variable for CV splits
Top N features	Number of top SHAP features to retain for downstream analysis

SHAP ranking — features ordered by mean absolute SHAP value
Model metrics — cross-validated performance (AUC for classification, R² for regression)
Cluster labels — sample subtype assignments (if clustering enabled)
Top feature table — candidates for follow-up validation (MetaSurv, orthogonal assays, manual review)

Survival analysis module for time-to-event endpoints.

Analysis	Description
Cox proportional hazards	Fits a regularised Cox model on omics features; reports C-index (cross-validated) and hazard ratios
Kaplan-Meier	Stratified KM curves for a grouping variable
Competing risk	Fine-Gray model for competing events (e.g. death from other causes)
Landmark analysis	Conditional survival from a landmark time point; corrects for guarantee-time bias

Parameter	Default	Description
Time column	—	Sample metadata column with follow-up time
Event column	—	Sample metadata column with event indicator (1 = event, 0 = censored)
CV folds	5	Cross-validation folds for Cox model evaluation